# Atlas AI Studio Overview

Most AI tools look incredible in demos. Then you try them in production and spend more time fixing outputs than you saved generating them. Atlas is different.

Atlas is AI-native production infrastructure for 3D content. It orchestrates 100+ AI models into game-ready pipelines for professional studios, sitting as the application layer between foundational AI models and shipping production work. Available on Google Cloud Marketplace.

## What Atlas does

Atlas pairs a node-based visual workflow editor with a multi-agent AI assistant. Artists and engineers compose production workflows from generation, optimization, and post-processing models without writing integration code, then run those workflows from inside Unreal Engine, Unity, Blender, or via direct API.

The platform spans the full 3D production pipeline: image generation, image editing, image-to-3D conversion, mesh optimization, retopology, UV unwrapping, PBR texturing, multi-view conditioning, video, audio, and animation. Each step is a node; nodes connect into reproducible, versioned, exportable workflows.

## Who Atlas is for

AAA and AA game studios that need to produce 3D content at scale without sacrificing production quality. Atlas is built for the customer that pays significantly more per asset than individual creators: studios with art directors, technical art teams, and engine programmers shipping real games on real deadlines.

Atlas customers include Square Enix, Parallel, and other AAA studios across Japan, Korea, Europe, and the United States.

## How to use this documentation

This guide explains how Atlas works, what each node does, and how to build production workflows. Start with Getting Started for a 10-minute orientation. Use Node Index for reference on every available capability. The User Manual covers operating the platform end to end. Atlas x GCP covers the Google Cloud deployment story.

The Ask feature on any page lets you query the docs directly. Append `?ask=<question>` to any URL and the server returns a focused answer with citations. Designed for AI agents and busy humans equally.

{% content-ref url="/pages/e6nB40Vt4vaQCg3prgSW" %}
[Getting Started](/atlas-ai-studio-overview/getting-started)
{% endcontent-ref %}

{% content-ref url="<https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md>" %}
<https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md>
{% endcontent-ref %}

{% content-ref url="/pages/bZW3OYUQix6hx1nTgPWw" %}
[3D & Gaming Focus](/atlas-ai-studio-overview/3d-and-gaming-focus)
{% endcontent-ref %}

{% content-ref url="/pages/WGxg5VROWJfkG7Hen0RT" %}
[Atlas x GCP](/atlas-ai-studio-overview/atlas-x-gcp)
{% endcontent-ref %}

{% content-ref url="/pages/nDq2E7XRFOeHp5U1wHuk" %}
[Testimonials](/atlas-ai-studio-overview/testimonials)
{% endcontent-ref %}

{% content-ref url="/pages/AXuWTHBqwv42iiXkm8So" %}
[User Manual](/atlas-ai-studio-overview/user-manual)
{% endcontent-ref %}

{% content-ref url="/pages/BTNe9YP7AV3m1QACbxBi" %}
[Node Index](/atlas-ai-studio-overview/node-index)
{% endcontent-ref %}


# Getting Started

Get started with Atlas. Covers the core ideas behind the platform: node-based visual workflow design, multi-model AI orchestration, asset-creation-specific AI agents, and integration with Unreal Engin

Atlas combines visual workflow design, diverse AI model orchestration, and asset creation specific AI agents into a single platform built for both artists and developers.

## Who Atlas is for

Atlas is designed for **professional game studios and creative teams** producing 3D content at scale. The platform is built around real production workflows rather than one-off asset generation, which is why most of its users sit in roles like:

* **Technical artists** building reusable asset pipelines for their studio
* **Art directors** maintaining style consistency across asset libraries
* **Engine programmers** integrating AI-generated content directly into Unreal Engine, Unity, or Blender
* **Production leads** measuring and reducing asset turnaround time
* **Studio CTOs and innovation teams** evaluating where AI fits in their pipeline

If your work involves shipping 3D assets to a game, a virtual production, or an interactive experience, Atlas is built for you.

**Visual Workflow Design**

A common criticism by artists using AI today is that it does not produce the result they want, and this can be seen as two separate issues:

1. Lack of control / editability
2. Lack of quality - reliance on a single AI model

Atlas uses a node based system which gives the artist a clear level of control to edit parts of assets in both 2D and 3D space in a non-destructive editor (point 1). And it encourages users to establish that control through the combination of AI models (e.g. isolating object parts with Gemini before generating them in 3D with CSM). Once an artist connects multiple AI models together the quality of the result improves significantly. Here is an example of an artist explaining his workflow:

> “I used images from Grok Imagine, refined them with Nano Banana, Tripo for raw 3D generation, Hunyuan for lowpoly and UV unwrap, Trellis2 for texturing and AI upscaling.”

Connect these generation and processing nodes on Atlas, and then add export nodes to create repeatable, version-controlled workflows that can be shared across teams or deployed as APIs - meaning the workflow runs at scale directly in Unreal Engine and Unity.

Sample Capabilities (AI Models available on Atlas)

* Text & Reasoning - Use AI extract key information from style guides and/or art bibles, and assist in prompting and workflow construction, and even automatically evaluate results based on user preferences.
* Image Generation - Generate concept art from descriptions and/or image references with multiple style options
* Multimodal Processing - Isolate objects, remove backgrounds, apply style transfer and edit everything from asset components to material properties
* Image-to-3D - Reconstruct detailed 3D meshes from single reference images with a diversity of topology and texture options
* Mesh Optimization - Auto-scale, set real-world dimensions, adjust pivots, reduce polycount, clean topology and all the other post-processing work required to make assets game ready for different engines

Integration-First Architecture Design

Simply being able to create high quality assets on the Atlas platform isn’t sufficient as it takes artists out of their established content creation and world building pipelines.

Every workflow built on Atlas can be exported as a production API with a single click. Integrate Atlas directly into existing pipelines, DCC tools, or game engines. Full versioning ensures deterministic, reproducible results and quality control.

* Direct integration with Unreal Engine, Unity, and Blender (and custom engines)
* Webhook support for async job processing
* Complete audit trail for enterprise compliance

*In practice, this simply means once a workflow has been built with Atlas the entire team can use it directly in their software of choice.*

## How to navigate this documentation

* [**Node Index**](/atlas-ai-studio-overview/node-index) — complete reference for every available node, with use cases, pitfalls, and FAQs per category. Start here when you're building or debugging a specific workflow.
* [**3D & Gaming Focus**](/atlas-ai-studio-overview/3d-and-gaming-focus) — what makes Atlas different in the 3D-and-gaming context (retopology, PBR, engine-specific presets, integration patterns).
* [**Atlas AI Agent**](https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md) — how the multi-agent assistant helps build workflows, diagnose issues, and propose improvements.
* [**Atlas x GCP**](/atlas-ai-studio-overview/atlas-x-gcp) — the Google Cloud deployment story, including procurement, billing, and integration with existing GCP commits.
* [**Testimonials**](/atlas-ai-studio-overview/testimonials) — real customer feedback from studios shipping with Atlas in production.

## Common starter pitfalls

A few mistakes that consistently slow down new users:

* **Treating Atlas like a single-model generator.** The platform's biggest leverage comes from chaining multiple models (e.g., image generation → 3D conversion → mesh optimization → texturing) into one workflow. Single-step usage misses most of the value.
* **Skipping the Atlas AI Agent.** The agent is trained to help with workflow construction; new users who try to manually wire every node spend hours where the agent would have taken minutes. Lean on it during early exploration.
* **Generating before defining the production target.** Atlas can produce wildly different output styles depending on backend and parameter selection. Pin the target (engine, polycount budget, texture resolution, style direction) before you start, not after.
* **Ignoring the API export.** Workflows that stay inside the Atlas web UI work for individuals, but the real production lift comes from exporting workflows as APIs and integrating them into existing tools (Unreal, Unity, Blender, custom pipelines). See [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) for the export flow.
* **Comparing single-shot Atlas output to Meshy or Tripo.** Atlas is orchestration across many backends including those. A like-for-like single-model comparison misses Atlas's defining capability. Compare Atlas pipelines to your current studio pipeline, not to single-model competitors.

## Frequently asked questions

**Do I need to know how to code to use Atlas?**

No. The platform is built for artists and technical artists, not developers. The node-based visual editor and the Atlas AI Agent both handle workflow construction without requiring scripting. Code becomes relevant only when integrating exported APIs into game engines or custom backends, and you can use the official [Unity and Unreal plugins](/atlas-ai-studio-overview/node-index/api-nodes) to skip that step too.

**Which game engines does Atlas integrate with?**

Atlas has official plugins for Unity (2023.1+) and Unreal Engine (5.5+). It also integrates with Blender via an add-on. Workflows exported as APIs work with any HTTP-capable engine or custom backend.

**How do I avoid spending too many credits during exploration?**

Use the lightweight versions of generation nodes during ideation (Text → Image (Fast), Fast Image → 3D), and reserve the high-quality variants for final passes. The Atlas AI Agent can also help structure workflows efficiently to avoid redundant generation steps.

**Can multiple team members collaborate on the same workflow?**

Yes. Workflows are version-controlled and shareable. Once a workflow is built and tested, it can be exported as an API and used by the entire team through their software of choice, or imported into the Atlas web UI for collaborative editing.

**What kind of input does Atlas need to produce high-quality output?**

Generation quality tracks input quality. For 3D generation specifically, clean reference images (isolated subject, neutral background, three-quarter angle) produce dramatically better results than cluttered inputs. See [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices) for the deeper guide.

**Is Atlas suitable for solo developers or only for AAA studios?**

The platform is designed for studios with production pipelines, but solo developers and small teams can use it too. The platform's strengths (multi-model orchestration, workflow reuse, API export) compound at studio scale, but the underlying capabilities work for any production size.


# Atlas AI Agent

The Atlas AI Agent is the multi-agent system that assembles game-ready workflows from natural-language instructions, hides hyperparameter complexity, and helps artists discover the right combination o

Underpinning the platform is a multi-agent system trained to help users build workflows, diagnose issues and propose improvements based on extensive 2D and 3D benchmarking not only research results, but practical out-of-distribution stress tests, and critically real-world artist usage. Every AI model we integrate into the platform gets tested against real tasks by real artists. We iterate based on what they actually ship.

Aside from allowing us to hide a majority of very specific and technical AI hyperparameters from the user (which the agent automatically sets), the agent is fundamentally a way to help someone unfamiliar with AI (like an artist) navigate and understand what these diverse models are capable of both in isolation and in combination.

**Key User Features**

* **Conversational Interface.** Describe what you need in plain language. Refine through dialogue. No prompt engineering required.
* **Context-Aware Generation.** The agent understands your project, style guidelines, and previous assets. Outputs stay consistent.
* **Real Workflow Training.** Trained on professional game art pipelines, knows the difference between concept and production as well as various art styles and direction.

**Key Technical Features**

* **Context engineering.** The agent decides what goes into the context window. You don't manually copy-paste code files and error logs. The app does the retrieval, embedding, and curation. This is a ton of hidden work.
* **Multi-call orchestration.** Under the hood, there are embedding models for your files, chat models for reasoning, models that apply nodes. The user sees one experience. The app runs a whole orchestra.
* **Application-specific GUI.** Text is hard to audit. Seeing nodes and connections uses your visual system, which is way faster than reading, and it gives the artist way more control.

<br>

## When to lean on the agent

The agent is most valuable in three situations:

* **First-time workflow construction.** Describe the production target in plain language (e.g., "Generate game-ready props from concept images at 5k polycount with PBR materials for Unreal Engine") and let the agent propose a starting workflow. Iterate from there.
* **Debugging unexpected output.** If a workflow is producing low-quality results, the agent can analyze the chain and propose specific node changes (different backend, different hyperparameter, additional pre or post-processing step). This is faster than trial-and-error on the user's part.
* **Discovering new capabilities.** The platform integrates a wide range of models and node types. The agent surfaces relevant capabilities the user might not know about, especially when crossing between domains (e.g., from image generation into 3D, or from a 3D mesh into animation).

For experienced users running a known workflow at scale, the agent is less essential. The biggest leverage is in early exploration and in any situation where the production target is shifting.

## Common pitfalls

* **Vague prompts produce vague workflows.** "Make a 3D model" is a weaker instruction than "Generate a 5,000-polycount stylized prop from this image with PBR materials and a tri-only topology for Unreal Engine." The richer the production target you communicate, the more precise the agent's output.
* **Skipping the agent and wiring nodes manually too early.** New users sometimes treat the agent as a beginner crutch and try to wire every node by hand. The agent is faster for almost all workflow construction, including for experienced users. Lean on it.
* **Ignoring the agent's debugging suggestions.** When a workflow produces unexpected output, the agent's proposed changes are usually based on benchmark patterns across many real production runs. Try its suggestions before reverting to manual fiddling.

## Related pages

* [Getting Started](/atlas-ai-studio-overview/getting-started) — broader orientation to the platform
* [3D & Gaming Focus](/atlas-ai-studio-overview/3d-and-gaming-focus) — the production-grade 3D context the agent operates inside
* [Node Index](/atlas-ai-studio-overview/node-index) — the full set of building blocks the agent can compose
* [Testimonials](/atlas-ai-studio-overview/testimonials) — customer quotes consistently highlight the agent as the differentiator

## Frequently asked questions

**Is the Atlas AI Agent a single model or a system?**

It is a multi-agent system, not a single model. Under the hood, different components handle context retrieval, reasoning over workflow construction, hyperparameter selection, and execution against the node graph. The user sees one conversational interface; the system runs multiple specialized agents.

**Can the agent build any workflow, or are there limits?**

The agent operates over the full set of nodes available in Atlas. If a node exists in the platform, the agent can incorporate it into a workflow. For capabilities not yet exposed as nodes, the agent cannot improvise.

**Does the agent require prompt engineering skills?**

No. The agent is built for artists and production teams, not prompt engineers. Clear descriptions of the production target (engine, polycount budget, texture resolution, style direction) are far more useful than carefully crafted prompts.

**How does the agent stay current as new models are added to Atlas?**

The agent's model coverage updates as new AI models are integrated into the platform. Every new model gets benchmarked against real tasks by real artists before it is exposed through the agent, so the agent's suggestions reflect production performance, not just research claims.

**Can the agent be used inside an exported API workflow, or is it only available in the web UI?**

The agent is the workflow-construction layer; once a workflow is built and tested, the exported API runs the workflow itself, not the agent. Workflows produced by the agent are first-class Atlas workflows and can be exported as APIs like any other.


# 3D & Gaming Focus

Atlas's 3D and gaming focus: retopology, PBR materials, UV unwrapping, multi-view texturing, engine-specific presets, and API integration with Unreal, Unity, and Blender. The production-grade capabili

2D generative models are very helpful for a diversity of use cases from level design to marketing video generation, but one thing that sets Atlas apart is a focus on 3D and gaming specific AI tools.

The platform integrates a diverse range of AI technologies all driven by gaming workflows, but the core focus (and where most of our proprietary technology sits) is in helping 3D generations move from simple prototypes to game ready assets.

This includes technologies and workflows that:

* Split objects into separate parts (pre and post generation) for higher 3D quality and bespoke editability
* Retopology controls that deliver optimization specific to different engines and control for manual editing
* Multi-view texturing and UV unwrapping capabilities to uprez, edit or even replace textures
* Full PBR material generation (including 8K+ textures)
* Automatic pivot, scale, and alignment fixes based on an understanding of real-world constraints and artist inputs
* Parallel and batch processing for multiple generations at once
* Engine-specific presets and automatic geometry / texture checks
* Integration of AI workflows directly into game dev pipelines (via API)

## Why a 3D-and-gaming focus matters

Generic AI 3D tools optimize for "make a 3D model from text or an image." That works for individual creators making one-off assets. It fails at the production scale studios need, because shipping a real game asset is not a single-step generation problem. It's a chain of decisions about topology, UVs, materials, scale, pivot, LODs, engine targets, and integration that an asset has to survive before it lands in a level.

Atlas treats every one of those steps as a first-class capability. The result is a platform that produces assets which actually ship into Unreal Engine or Unity, not just demos that look good in a presentation.

## Use cases by studio function

* **Technical artists** use Atlas to build reusable asset pipelines: image-to-3D with the right backend, automatic retopology to a target polycount, baked normal maps, engine-specific export presets. Once built, the same workflow runs on every new asset.
* **Art directors** use Atlas to maintain style consistency across asset libraries: a shared reference image plus style-conditioning means every new asset matches the established direction.
* **Engine programmers** use the exported API to trigger generation from inside Unreal Engine or Unity, often via the [official plugins](/atlas-ai-studio-overview/node-index/api-nodes). Players or designers request assets; the engine receives game-ready GLB output.
* **Production leads** use parallel and batch processing to compress timelines. A workflow that handles a single asset in 30 seconds can run 100 in parallel for asset library expansion.
* **Studio CTOs and innovation teams** use Atlas to evaluate AI integration with concrete production metrics, not generic claims. The integration-first architecture means Atlas slots into existing pipelines rather than replacing them.

## Production patterns worth knowing

* **Generate → optimize → bake → texture.** The canonical chain for hero assets. Generate at high detail, retopologize to a low-poly target, bake high-detail features as normal and PBR maps onto the low-poly mesh. See [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices) for the deep dive.
* **Multi-view conditioning for character work.** Single-image generation hallucinates backs and occluded surfaces. For characters, use Multi-View → 3D from [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) with front, side, and back references.
* **Style-locked batch generation.** Use a canonical reference image extracted via Describe Image (in [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes)) to lock style, then batch-generate variants of a hero asset that share visual direction.
* **Engine-aware export.** Atlas's automatic geometry and texture checks catch issues before they reach the engine. Combined with the engine-specific presets, this is the difference between assets that import cleanly and assets that need manual fixes.

## Related pages

* [Node Index](/atlas-ai-studio-overview/node-index) — complete reference for every node, organized by category
* [Atlas AI Agent](https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md) — how the multi-agent system helps build production workflows
* [Atlas x GCP](/atlas-ai-studio-overview/atlas-x-gcp) — the Google Cloud Marketplace deployment story
* [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices) — the depth-first guide to 3D generation
* [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) — exporting workflows as callable APIs for engine integration

## Frequently asked questions

**How is Atlas different from Meshy, Tripo, or other AI 3D generators?**

Atlas is an orchestration layer that runs many models, including Meshy, Tripo, Hunyuan, Trellis, and others, as backends. The platform's value is in the production-grade workflow built around the model layer: retopology, UV unwrapping, PBR generation, engine export, integration plugins. Single-model generators produce raw output; Atlas produces shippable assets.

**Which game engines does Atlas target?**

Unreal Engine 5.5+, Unity 2023.1+, and Blender via official plugins. Custom engines and other tools integrate via API. The platform is engine-agnostic by design; the export format (GLB) and the API surface work with anything HTTP-capable.

**Does Atlas support PBR materials at production resolution?**

Yes. The platform supports full PBR material generation including 8K+ texture resolutions. Texture size and resolution are exposed as workflow parameters, so studios can dial in the right level of detail for the target engine and platform.

**How does retopology work in Atlas?**

Retopology is a built-in capability via the [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes). Options include target polycount, target topology (triangles or quads), and engine-specific presets that respect the polycount and UV layout conventions of each target engine.

**Can Atlas produce game-ready characters with rigs and animations?**

Yes. The full pipeline is: Multi-View → 3D for the character mesh, Rig Humanoid Mesh for skeleton, Retarget Rigged Mesh and Animate Rigged Model for motion. The Character Workflow node chains all of these into a single automated step. See [Animation Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes/animation-nodes) for the full guide.

**What's the typical asset turnaround time?**

Depends on the workflow. Concept-to-game-ready for a single asset typically takes 1 to 5 minutes end-to-end, compared to days or weeks for traditional pipelines. Hero assets that require extensive baking and refinement take longer.


# Atlas x GCP

Atlas is a Google Cloud Partner available exclusively via Google Cloud Marketplace. Covers the procurement, billing, and integration advantages of the GCP-native deployment for studios in the era of A

Atlas is an agentic 3D-content creation platform designed for professional game studios, enabling them to generate game-ready assets, environments, tools, and workflows. It focuses on production-scale workflows rather than one-off asset generation, acting as a creative assistant through its multi-agent AI system. Developers can co-create with intelligent AI agents using natural language prompts, ensuring the output is tailored to their specific technical and aesthetic goals. Atlas integrates with industry-standard pipelines like Unreal Engine, Unity, and Houdini, and embraces the idea that we are entering a new era of AI-native game development.

**The Opportunity - A New Era of AI-Native Game Development**

The games industry is on a powerful ride, surging forward with innovation and a sharp focus on the player experience. For years, the industry’s evolution was defined by familiar IPs getting better graphics and gameplay. We believe we’re on the cusp of something far more radical — a shift on the scale of the transition from cartridges to CD-ROMs, or 2D to 3D graphics. This new era is defined by the rise of "living games," a new form of dynamic, ever-evolving experiences powered by AI that captivate players for years.

With the global market for games surpassing $180 billion in 2024, this fundamental shift in how games are developed, played, and experienced creates an entirely new opportunity for the industry.

**The Problem**

The process of game asset creation fundamentally hasn't changed in 20 years.

> concept art creation → modeling → UV mapping → texturing → optimization → export

This painful path requires specialized talent, takes days, and unsustainably expensive in an era where players demand larger maps, higher fidelity assets, and more immersive experiences.\
\
The Result:

* Unsustainable costs
* Production bottlenecks / delays
* Talent scarcity
* Zero scalability / flexibility

**The Solution - Atlas AI Studio x GCP**

Atlas has developed an end-to-end AI platform that allows studios to create specific workflows that deliver on idea to game-ready 3D asset in minutes - through an interface artists actually want to use (non-destructive and integrated into existing pipelines).

The fact that the Atlas platform is **exclusively** available via GCP brings several important advantages:

1. Customer can optimize cloud spend by applying existing commits toward Atlas
2. Faster procurement with pre-negotiated terms, no new vendor paperwork, and enterprise grade security & IP protections
3. Unified billing - single invoice through Google Cloud.

## When the GCP Marketplace path makes sense

For most studios already operating on Google Cloud, the Marketplace path is the fastest way to onboard Atlas:

* **Existing GCP customers.** If your studio already has a GCP commit or an enterprise agreement, applying Atlas to that commit is the lowest-friction onboarding. No new vendor relationship, no fresh procurement cycle.
* **Studios with strict procurement requirements.** AAA studios often have months-long vendor onboarding processes. The GCP Marketplace bypasses most of this because Google has pre-negotiated terms covering security, IP, and contract structure.
* **Studios standardizing on cloud-native tooling.** Atlas runs on GCP-native infrastructure. For studios already operating cloud-native production pipelines on Google Cloud, Atlas fits the existing operational pattern rather than introducing new surface area.
* **Cross-studio organizations.** A single GCP Marketplace contract can cover multiple studios within a parent organization, simplifying procurement at the parent level rather than studio-by-studio.

## What integration with GCP means in practice

Atlas's exclusive GCP availability is more than a procurement convenience. It has direct operational implications:

* **GCP-native infrastructure.** Atlas runs on Google Cloud infrastructure, which means scaling and operational characteristics line up with the rest of your GCP-native workloads.
* **Marketplace billing reconciliation.** Atlas usage appears alongside the rest of your cloud spend on a single Google Cloud invoice. Finance teams reconcile one invoice rather than two.
* **Co-marketing opportunities with Google Cloud.** Atlas customers are eligible for joint case studies and visibility under the Google Cloud gaming program. Atlas is already featured in the Google Cloud "A New Era of Gaming" coverage.
* **Enterprise security and IP protection.** The Marketplace listing includes the enterprise-grade security and IP protections that come with the standard Google Cloud contract structure, which most studio legal teams already accept.

## Related pages

* [Getting Started](/atlas-ai-studio-overview/getting-started) — orientation to the Atlas platform itself
* [3D & Gaming Focus](/atlas-ai-studio-overview/3d-and-gaming-focus) — what Atlas does differently for the 3D-and-gaming context
* [Atlas AI Agent](https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md) — the multi-agent assistant that drives workflow construction
* [Testimonials](/atlas-ai-studio-overview/testimonials) — feedback from studios shipping with Atlas
* [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) — exporting workflows as production APIs for engine integration

## Frequently asked questions

**Is Atlas only available on Google Cloud?**

Yes. Atlas is exclusively available via Google Cloud Marketplace. The platform is built on GCP infrastructure and the deployment model is not portable to AWS or Azure.

**Can I use my existing GCP commit toward Atlas?**

Yes. This is one of the primary advantages of the Marketplace listing. Atlas usage applies against existing GCP commits and enterprise agreements, which means Atlas spend can come out of capacity you've already committed to Google Cloud.

**How does procurement work for AAA studios with long vendor approval cycles?**

The GCP Marketplace path is typically the fastest. Studios already approved to purchase via Google Cloud do not need to onboard Atlas as a separate vendor. The pre-negotiated terms include enterprise-grade security, IP protection, and contract structure that most legal and IT teams accept without further review.

**Does the GCP integration affect Atlas's API output or workflows?**

No. The workflows, the node-based editor, the AI Agent, and the API export all behave identically regardless of how you procured Atlas. The GCP path is about procurement and billing, not about platform capability. Workflows exported as APIs work with Unity, Unreal, Blender, and custom backends running on any cloud or on-premises.

**Is there a public case study on the Atlas-and-GCP partnership?**

Atlas is featured in Google Cloud's "A New Era of Gaming" blog. The published coverage is a useful starting reference for procurement conversations with finance and IT teams.

**What about studios not currently on Google Cloud?**

Studios new to Google Cloud can adopt Atlas through standard Marketplace onboarding. The GCP customer success team typically supports new-customer onboarding, including the architecture and security review process.


# Testimonials

Testimonials

Atlas is trusted by innovative studios building the next generation of games and interactive experiences.

> "Atlas's ability to seamlessly integrate with our highly customized workflows has been a game-changer. By deeply understanding the nuances of our pipeline, they've become an invaluable partner, enabling us to deliver high-quality, performance-optimized solutions with impressive agility."\
> Tech Director, Innovation Technology Division, Square Enix

> "Atlas has completely changed our production timelines. Assets that used to take one to two weeks are now ready in a day or two, without compromising quality. It fits naturally into how our team already works, which made adoption incredibly smooth."\
> Production Lead, AAA Gaming Studio

> “Using Atlas, we were able to automatically remaster 20,000 existing assets with consistency and speed. Doing this manually would have been unrealistic at best. It unlocked value from our library that we simply could not access before.”\
> Art Director, AA Mobile Gaming Studio

Proven Results from Beta Testing phase

* 10-50x faster asset creation based off internal testing with 5 AAA design partners working on environment assets and props.
* 70-90% cost reduction based off AA game production for Alpha Release.
* 5-10x artist productivity gain based off user feedback from 20+ design partners using the platform for various game asset productions.

## Related pages

* [Getting Started](/atlas-ai-studio-overview/getting-started) — orientation to the Atlas platform
* [3D & Gaming Focus](/atlas-ai-studio-overview/3d-and-gaming-focus) — the production-grade capabilities customers reference in the quotes above
* [Atlas AI Agent](https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md) — the multi-agent system that customers consistently call out as the differentiator
* [Atlas x GCP](/atlas-ai-studio-overview/atlas-x-gcp) — how studios procure Atlas through the Google Cloud Marketplace


# User Manual

The Atlas User Manual is in active development. In the meantime, this page points to the existing documentation that covers each major area of the platform: node reference, agent guidance, gaming-spec

The full Atlas User Manual is in active development. Most of the material that would sit in a traditional user manual already lives in topic-specific pages elsewhere in this documentation. Use this page as the index until the unified manual ships.

## Where to find what you need

* **Building your first workflow.** Start with [Getting Started](/atlas-ai-studio-overview/getting-started) for the orientation, then open the [Node Index](/atlas-ai-studio-overview/node-index) to browse the building blocks. The [Atlas AI Agent](https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md) can also generate a first-pass workflow from a natural-language description.
* **Picking the right node for a task.** The [Node Index](/atlas-ai-studio-overview/node-index) is organized by capability area (mesh, image, video, text, utility, API, math, integration). Every node category has its own page with use cases, common pitfalls, and FAQs.
* **3D-specific guidance.** The [3D & Gaming Focus](/atlas-ai-studio-overview/3d-and-gaming-focus) page is the single best entry point for the production-grade 3D capabilities (retopology, PBR, UV unwrapping, engine presets). For deeper guidance on a single asset, see [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices).
* **Character pipelines specifically.** Character work (rigging, retargeting, animation) is covered in [Animation Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes/animation-nodes).
* **Exporting workflows as APIs.** [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) covers the one-click API export, the async execute and status pattern, and the official Unity and Unreal plugins.
* **Procurement and billing through Google Cloud.** [Atlas x GCP](/atlas-ai-studio-overview/atlas-x-gcp) covers the Marketplace path, including how Atlas usage applies against existing GCP commits.
* **What other studios are saying.** [Testimonials](/atlas-ai-studio-overview/testimonials) for quotes and measured results from beta and production users.

## What is coming in the full User Manual

The unified manual will consolidate the above into a single sequenced walkthrough, plus add deeper coverage of:

* Project organization, asset library structure, and shared workflow management within a team
* Permissions, roles, and audit trail for enterprise deployments
* Integration recipes for specific engine workflows (Unreal blueprints, Unity scriptable objects, Blender add-ons)
* Custom node development for studios with proprietary models or in-house tooling

In the meantime, the topic-specific pages above are the source of truth. The [Atlas AI Agent](https://github.com/Atlas-Design/docs/tree/main/atlas-ai-studio-overview/README.md) can also answer most "how do I" questions interactively from inside the platform.

## Frequently asked questions

**Where is the actual user manual?**

The full unified manual is in development. The topic-specific pages linked above cover all current platform capabilities. The agent inside the Atlas platform also answers most usage questions interactively.

**Can I get a single-document export of the documentation?**

Not currently. The documentation is structured page-by-page in GitBook. Specific pages can be saved or printed individually.

**Who do I contact if I cannot find what I need?**

Atlas customers have a dedicated point of contact assigned at onboarding. Prospective users can reach the team via the contact form linked from atlas.design.


# Node Index

Complete reference for every node available in Atlas workflows. Covers input, image, mesh, video, audio, API, and utility nodes with use cases, examples, and best practices for production-grade 3D con

Atlas workflows are built from nodes. Each node is a discrete capability: a model that generates content, a transform that processes it, a utility that routes data, or an API surface that exposes the workflow to external tools. Connect nodes together and you get a reusable production pipeline.

This index covers every node category on the Atlas platform. Use it as a reference when you're building or troubleshooting a workflow, or as a starting point for understanding what's possible. Each subsection includes node descriptions, use cases, common pitfalls, and frequently asked questions.

## Node categories

* **Input Nodes** — bring text, images, references, PDFs, and other source content into a workflow
* **Image Nodes** — generate, edit, and post-process 2D images at production quality
* **Mesh Nodes** — generate, refine, optimize, and prepare 3D meshes for game engines
* **Video Nodes** — generate and edit short video content from prompts and references
* **Audio Nodes** — synthesize speech, music, and sound effects from text or other inputs
* **API Nodes** — export workflows as callable REST endpoints for Unity, Unreal, Blender, or any backend
* **Utility Nodes** — orchestrate data, manipulate arrays, run LLM operations, structure outputs

{% content-ref url="/pages/z5pWvVdsIs9yrkRBnYBA" %}
[Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes)
{% endcontent-ref %}

{% content-ref url="/pages/b4tN1RB4hbAu9ngELKqI" %}
[Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes)
{% endcontent-ref %}

{% content-ref url="/pages/X47xyoIuTIP5yF2M7p8D" %}
[Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes)
{% endcontent-ref %}

{% content-ref url="/pages/MRJmZU0Oi1Apux53AqG6" %}
[Video Nodes](/atlas-ai-studio-overview/node-index/video-nodes)
{% endcontent-ref %}

{% content-ref url="/pages/GfrKAa6nxmtPP6BDC8mn" %}
[API Nodes](/atlas-ai-studio-overview/node-index/api-nodes)
{% endcontent-ref %}

{% content-ref url="/pages/5TBpdTmEC1N31vwUvij7" %}
[Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes)
{% endcontent-ref %}


# Input Nodes

Input nodes bring images, text, meshes, audio, video, PDFs, EXR, USDZ, and parameters into Atlas AI Studio workflows for AI generation.

Input Nodes are how source content enters an Atlas workflow. Every pipeline starts with one or more input nodes: a reference image, a text prompt, a parameter value, a PDF design document, a 3D mesh to refine, or an audio clip to process. The platform supports inputs across every content type Atlas operates on.

## When to use input nodes

Use input nodes whenever you need to pass material from outside the workflow into the graph. Common patterns:

* **Reference-driven generation.** Drop a concept art image into an Input Image node to drive style or composition in downstream generation nodes.
* **Parameter exposure.** Use Input Number or Input Boolean nodes to expose tunable settings (seeds, dimensions, toggles) that the workflow operator can adjust per run.
* **Document-driven workflows.** Feed game design documents (GDDs) or art bibles via Input PDF, then extract text and images downstream for use in generation prompts.
* **3D refinement.** Bring an existing mesh into the workflow via Input Mesh to retopologize, retexture, or rig it without regenerating geometry.
* **Animation and motion.** Use Input Mixamo Animation to bring FBX animation clips for retargeting onto generated characters.

### Input Boolean

* Lets you manually toggle a simple Yes/No (True/False) value.

<figure><img src="/files/ol4pSf4t7kFl1b69PZC6" alt="" width="563"><figcaption></figcaption></figure>

### Input Number

* Provides a way to manually input / control specific numerical values, for example the 'seed' of an AI generation.

<figure><img src="/files/hDucApIOMmejaP1D10Yk" alt="" width="563"><figcaption></figcaption></figure>

### Input Text

* Allows you to manually enter text strings for use in your graph. Most AI models on the platform have reasoning capabilities and are therefore able to handle extensive and or complex prompts.

<figure><img src="/files/hXbZJzBM5ls91FlycqdO" alt="" width="563"><figcaption></figcaption></figure>

### Input Image

* Used for manually uploading or selecting a single image file from your device.

<figure><img src="/files/eLj3tIYuM0QTt8QeiehS" alt="" width="563"><figcaption></figcaption></figure>

### Input Images

* Allows for manual uploading and management of a collection of multiple images treated as a list.

### Input Mesh

* Enables the manual upload or selection of 3D model (mesh) files in GLB format.

<figure><img src="/files/9CwCg8e0iGI0W8Bv3Piq" alt="" width="563"><figcaption></figcaption></figure>

### Input PDF

* Provides a field to upload and process PDF documents.

<figure><img src="/files/G9zmiOcjHWn9kyu0cd3b" alt="" width="563"><figcaption></figcaption></figure>

### Input Video

* Allows you to upload or select video files from your computer to add to your pipeline.

<figure><img src="/files/7jH1oXPqm1z3GuFOwWth" alt="" width="563"><figcaption></figcaption></figure>

### Input Audio

Supplies audio files into Atlas workflows for processing, analysis, or transformation. Accepts standard audio formats and passes the file to downstream nodes that support audio input.

<figure><img src="/files/poJKw0evaOQZpUSClSqr" alt="" width="563"><figcaption></figcaption></figure>

* Upload field accepts common audio formats (MP3, WAV, OGG, FLAC).
* Output provides the audio file reference for connection to audio-processing or transcription nodes.
* File size and duration limits depend on the backend receiving the audio.
* Retains original audio properties (sample rate, channels, bit depth) unless modified by downstream processing.
* Useful for providing voice-over recordings, background music stems, or sound effect samples for analysis, transcription, or AI-driven remixing in game audio pipelines.

### Input Mixamo Animation

Supplies pre-downloaded animation files from the Mixamo library into Atlas workflows for rigging, retargeting, or motion editing. Designed to work with FBX animation data exported from Mixamo's web platform.

<figure><img src="/files/MBuXD1sgoJ9cTkdr0HYQ" alt="" width="563"><figcaption></figcaption></figure>

* Upload field accepts FBX files containing skeletal animation data.
* Output provides the animation reference for connection to retargeting, blending, or character rigging nodes.
* Does not fetch animations directly from Mixamo; files must be downloaded externally and uploaded through this node.
* Preserves animation clip properties including keyframe timing, bone hierarchy, and motion curves.
* Compatible with humanoid rigs and generic skeletal structures depending on the downstream retargeting or processing node.
* Useful for rapidly prototyping NPC locomotion cycles, combat animations, or cutscene performances using Mixamo's motion-capture library without manual keyframing.

### Input PDFs

Passes one or more PDF documents into the workflow for downstream text and image extraction.

* **Multiple file inputs.** Accepts several PDF files simultaneously within a single node.
* **Document source.** Brings external documents such as game design documents (GDDs), art bibles, or lore references into the graph.
* **Downstream extraction.** Feeds connected nodes that pull text or embedded images from the supplied PDFs.
* **Batch handling.** Groups related documents together so they can be processed in one pass.
* **Output.** Provides the loaded PDF collection for use by extraction and generation nodes.

<figure><img src="/files/ccT3pfdtIaHrLrTy3Gj5" alt="" width="563"><figcaption></figcaption></figure>

Useful for: driving prompts and asset generation from written design references, extracting narrative or lore text for NPC dialogue, and consolidating multiple documents into a single workflow input.

### Input EXR

Supplies an EXR image file into a workflow for use by downstream nodes. EXR files carry high dynamic range and multi-channel data, making them suited to lighting references, HDR environments, and layered render passes.

<figure><img src="/files/jY2ApBIWClsHziPJ0Jmu" alt="" width="563"><figcaption></figcaption></figure>

* **Input:** an EXR file loaded into the node.
* **Output:** the EXR image data passed to connected nodes.
* **High dynamic range.** Preserves floating-point pixel values beyond the standard 0–1 range for accurate lighting and exposure information.
* **Multi-channel support.** Carries additional channels such as depth or auxiliary render passes where present.
* **Typical use.** Feed HDR environment maps or lighting references into generation and compositing nodes.

Useful for: driving scene lighting, environment backdrops, and high-fidelity reference imagery for game asset generation.

### Input USDZ

Supplies a USDZ file into a workflow, making 3D scene and model data available to downstream nodes.

<figure><img src="/files/Xe3283ppDv6sJT759gwF" alt="" width="563"><figcaption></figcaption></figure>

* **Input:** A USDZ file supplied to the workflow.
* **Output:** The USDZ asset passed to connected downstream nodes.
* **Format handling.** Accepts USDZ packages containing geometry, materials, and scene structure.
* **Refinement pipelines.** Feed existing 3D assets into the graph for retexturing, conversion, or inspection without regenerating geometry.
* **Asset reuse.** Bring authored props, characters, or environment pieces into a workflow to drive variation or further processing.

Useful for: importing prebuilt 3D assets into workflows for game environment and prop pipelines.

### Input Maximo Animation

Supplies a downloaded animation clip into the workflow for retargeting onto generated or imported characters.

* **Input.** Accepts an FBX animation file sourced from a Mixamo download.
* **Output.** Passes the animation clip downstream to rigging and retargeting nodes.
* **Skeleton-driven motion.** Provides the joint hierarchy and keyframe data needed to drive a target rig.
* **Reusable clips.** Bring in walk cycles, idles, attacks, or gestures without authoring motion inside the graph.
* **Typical use.** Pair with a generated character mesh and a retargeting node to apply prebuilt motion to new assets.

<figure><img src="/files/MBuXD1sgoJ9cTkdr0HYQ" alt="" width="563"><figcaption></figcaption></figure>

Useful for: prototyping NPC locomotion, populating cutscenes with prebuilt gestures, and applying stock animation clips across multiple generated characters.

## Common pitfalls

* **Uploading an unsupported file format.** Each input type accepts specific formats: Input Mesh expects GLB, Input Mixamo Animation expects FBX skeletal animation data, Input Audio accepts MP3, WAV, OGG, and FLAC, Input EXR expects EXR files, and Input USDZ expects USDZ packages. Mismatched formats produce silent failures in downstream nodes rather than an upload error.
* **Hard-coding values that callers need to vary.** When a workflow is exported as an API, constants set inside Input Text, Input Number, or other input nodes become unchangeable at execution time. Expose each variable value through its own input node so it can be replaced by the corresponding API Input node.
* **Chaining single Input Image nodes for multi-image inputs.** Multi-view 3D generation and batch processing nodes expect an array. Use Input Images, or build an array via Create Image List in [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes), instead of wiring several Input Image nodes together.
* **Confusing Input PDF with Input PDFs.** Input PDF processes one document; Input PDFs accepts several files at once for batch text and image extraction. Using Input PDF when consolidating multiple design references processes only a single document.

## Related nodes

* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — array management, text processing, and document extraction. Use these to transform inputs, including PDF text and image extraction, before feeding generation nodes.
* [Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes) — generation and editing nodes that consume Input Image, Input Images, or Input EXR.
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — 3D generation and processing nodes that consume Input Image for image-to-3D, or Input Mesh and Input USDZ for refinement and reuse of existing assets.
* [Video Nodes](/atlas-ai-studio-overview/node-index/video-nodes) — video generation and processing nodes that pair with Input Video.
* [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) — the external form of input nodes. Replace Input nodes with API Input nodes when preparing a workflow for external execution.

## Frequently asked questions

**What is the difference between Input Image and Input Images?**

Input Image accepts a single image file selected or uploaded from your device. Input Images accepts a collection (array) of multiple images treated as a list. Use Input Images when feeding multi-view 3D nodes, batch processors, or comparison workflows.

**Which 3D file formats can I upload?**

Input Mesh accepts 3D model files in GLB format. Input USDZ accepts USDZ packages containing geometry, materials, and scene structure. Convert other formats to a supported format before uploading if the node you are targeting does not accept them.

**Can I upload videos directly?**

Yes, via the Input Video node, which uploads or selects video files from your computer. Standard video formats are supported. For a text to video AI node and other generation-side capabilities, see [Video Nodes](/atlas-ai-studio-overview/node-index/video-nodes).

**Can I upload audio files?**

Yes. Input Audio accepts common audio formats including MP3, WAV, OGG, and FLAC, and passes the file reference to downstream audio-processing or transcription nodes. It retains original audio properties such as sample rate, channels, and bit depth unless modified by downstream processing. Use it to supply voice-over recordings, music stems, or sound effects for analysis, transcription, or lipsync generation in game audio pipelines.

**How do I supply high dynamic range or multi-channel images?**

Use Input EXR. EXR files carry high dynamic range and multi-channel data, preserving floating-point pixel values beyond the standard 0–1 range and additional channels such as depth or auxiliary render passes where present. Feed HDR environment maps or lighting references into generation and compositing nodes.

**How do I bring existing 3D scenes into a workflow?**

Use Input USDZ to supply a USDZ package containing geometry, materials, and scene structure to downstream nodes. This allows you to reuse authored props, characters, or environment pieces for retexturing, conversion, or inspection without regenerating geometry.

**How do I process multiple documents at once?**

Use Input PDFs to pass several PDF files simultaneously within a single node for downstream text and image extraction. It brings external documents such as game design documents, art bibles, or lore references into the graph and groups them for processing in one pass. Use Input PDF when working with a single document.

**How do I expose input values as parameters when exporting a workflow as an API?**

Replace each Input node with the corresponding API Input node before exporting. For example, Input Text becomes API Input Text. API Input nodes let external callers send values into the workflow at execution time. See [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) for the full export process.

**Can I fetch Mixamo animations directly from the Mixamo platform?**

No. The Input Maximo Animation node accepts FBX files containing skeletal animation data exported from Mixamo's web platform; it does not fetch animations directly. Download from Mixamo first, then upload. It preserves keyframe timing, bone hierarchy, and motion curves for retargeting, blending, or character rigging nodes.

**What is the file size limit for inputs?**

File size and duration limits depend on the backend receiving the input and the downstream node consuming it. If an upload fails, a common cause is file size; reduce resolution or polycount upstream before retrying.


# Image Nodes

Image Nodes in Atlas cover the full 2D production lifecycle: generating images from text or references, editing existing images with text instructions, and post-processing for framing, upscaling, and

Image Nodes are how 2D content gets created and refined in Atlas. They cover the full lifecycle: generating images from text prompts or references, editing existing images with text-driven instructions, and post-processing for camera framing, upscaling, background removal, and compositing.

In a typical Atlas workflow, image nodes do early-stage work: concepting, ideation, reference preparation, and 2D asset finalization. They also feed directly into [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) for 3D generation, which is why the input quality at this stage materially affects 3D output quality downstream.

## The three image-node categories

* **Image Generation Nodes** — text-to-image and image-to-text. Use these to create new images from prompts or extract text descriptions from existing images.
* **Image Edit Nodes** — text-guided image transformation. Modify existing images with natural-language instructions, with options ranging from fully generative edits to precise localized modifications.
* **2D Post Processing Nodes** — refinement and finishing. Camera framing, outpainting, upscaling, background removal, segmentation, and image composition.

{% content-ref url="/pages/T3sC9CB573vFi0hT5fgl" %}
[Image Generation Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-generation-nodes)
{% endcontent-ref %}

{% content-ref url="/pages/mjGTYmwRooWhuWUX6xlm" %}
[Image Edit Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-edit-nodes)
{% endcontent-ref %}

{% content-ref url="/pages/Von7MQyy6JP2gohAmnpH" %}
[2D Post Processing Nodes](/atlas-ai-studio-overview/node-index/image-nodes/2d-post-processing-nodes)
{% endcontent-ref %}


# Image Generation Nodes

Atlas AI Studio image generation nodes: text-to-image generation, image-to-text description, and procedural texture generation for game asset workflows.

This section introduces the core nodes used for **Text → Image** generation and **Image → Text** extraction.

## When to use image generation nodes

* **Concept exploration.** Use Text → Image (Fast) for rapid ideation, thumbnail generation, and quick variations during pre-production. Spending minimal credits to iterate on direction.
* **Production-quality concept art.** Use Text → Image (High Quality) when the output will be referenced by downstream nodes (image-to-3D, video, retexturing) or shown to stakeholders.
* **Reference description extraction.** Use Describe Image (image-to-text) to convert visual references into structured prompts. The output text feeds into other generation nodes for consistent style transfer.
* **Procedural textures and patterns.** Use Procedural Image Builder for tileable textures, repeating patterns, and parametric variations across material libraries.
* **Asset library generation.** Pair generation nodes with Create Image List from [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) to produce families of variants in a single workflow run.

<figure><img src="/files/VFQ68M9j3VZ1ISB1qTOT" alt="" width="563"><figcaption></figcaption></figure>

***

### **Text → Image** (Fast) <a href="#fast-text-to-image" id="fast-text-to-image"></a>

The **Text → Image (Fast)** node generates images quickly based on a text prompt.\
It uses a lightweight backend optimized for **speed** rather than fidelity.

This node is ideal when you need:

* rapid ideation
* quick thumbnails
* early explorations of a concept
* instant variations

The output is less detailed and less accurate compared to the **High Quality** version of the node, but significantly faster.

<figure><img src="/files/U7uIxx4CC0GhWsEpSeTb" alt="" width="563"><figcaption></figcaption></figure>

***

### Text **→** Image (High Quality) Node <a href="#quality-text-to-image" id="quality-text-to-image"></a>

The **Text → Image (High Quality)** node produces **high-fidelity, detailed images**.\
It uses **different backends**, allowing you to choose based on rendering styles and strengths.

#### Key Features <a href="#key-features" id="key-features"></a>

* best visual quality among text-only generation nodes
* multiple backends → diverse interpretations
* good for environments, props, characters, concept art
* consistent structure and lighting compared to faster models

#### Example: Fast v. High Quality <a href="#example-quality-vs-fast" id="example-quality-vs-fast"></a>

Both nodes receive the same prompt.

The Fast node generates a quick idea; the High Quality node generates a polished version suitable for downstream workflows such as:

* Image → 3D
* Text + Image → Image
* Edit Image with Text
* Marketing assets

<figure><img src="/files/nhwgW7NDTXYsbCnFulb8" alt="" width="563"><figcaption></figcaption></figure>

***

### Describe Image Node <a href="#image-to-text" id="image-to-text"></a>

The **Describe Image** node converts an image into structured or descriptive text. The system prompt controls what type of information is extracted.

These nodes can be powerful ways to craft custom, detailed prompts for image generation.

#### You can request: <a href="#you-can-request" id="you-can-request"></a>

* scene description
* object list
* spatial layout
* material analysis
* asset identification
* style breakdown
* inferred function or purpose

<figure><img src="/files/cpZO0rrH7ZbqD0JXlKLh" alt="" width="563"><figcaption></figcaption></figure>

#### Example Use Cases <a href="#example-use-cases" id="example-use-cases"></a>

* Describe the objects in a rendered scene
* Extract the composition of a top-down plan
* Identify modular parts of a building
* Produce a clean text prompt for concept art
* Provide metadata for Make3D or Auto Mesh Transform

### Describe Images Node

The multi-image variant of Describe Image: converts multiple images into structured or descriptive text, with the same request types and system-prompt control described above.

### Procedural Image Builder

Generates textures and materials using algorithmic patterns rather than AI synthesis. Creates geometric, noise-based, or mathematical designs suitable for tiling backgrounds, UI elements, and placeholder art.

<figure><img src="/files/WPkwSGwgLVK9DyKhYk2S" alt="" width="563"><figcaption></figcaption></figure>

* **Pattern types** — Voronoi, Perlin noise, grid, checkerboard, and other mathematically defined textures.
* **Output resolution** — Configurable width and height for the generated image.
* **Tileable option** — Ensures seamless wrapping for repeated textures across surfaces.
* **Parameter inputs** — Scale, seed, color palette, and pattern-specific controls adjust the visual result.
* **Deterministic** — Identical parameters always produce the same output; no randomness unless seed is changed.
* **Backend selector** — Some backends offer extended pattern libraries or performance optimizations.

**Useful for:** rapid prototyping of environmental textures, UI backgrounds, stylized materials for low-poly assets, and placeholder art during early production phases.

* **Code-generated output** — Produces images through generated code that constructs patterns, gradients, and compositions.

## Common pitfalls

* **Shipping Fast output as production assets.** The Text → Image (Fast) node produces visibly lower detail and accuracy than the Text → Image (High Quality) node. Using it for final assets yields results unsuitable for downstream work such as Image → 3D.
* **Picking a High Quality backend without testing.** The Text → Image (High Quality) node exposes multiple selectable backends with distinct rendering styles. Choosing one that does not match the intended style direction wastes iterations; test a few with the same prompt early to lock one in.
* **Re-rolling instead of editing.** Once a generation is close, re-running the whole prompt discards the parts that already worked. Refine specific aspects with [Image Edit Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-edit-nodes) for faster, more controlled iteration.
* **Asking a text-to-image node for seamless textures.** Text-to-image nodes do not guarantee wrapping edges, so repeated results show visible seams. Use the Procedural Image Builder node with the Tileable option for repeating textures.

## Related nodes

* [Image Edit Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-edit-nodes) — refine and modify images generated here without re-rolling the whole prompt.
* [2D Post Processing Nodes](/atlas-ai-studio-overview/node-index/image-nodes/2d-post-processing-nodes) — camera framing, upscaling, background removal, and outpainting for finalizing generated images.
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — image-to-3D nodes consume the output of image generation nodes; image quality at this stage affects 3D output.
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — Create Image List for batching and Text Generation (LLM) for prompt refinement.
* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — Input Text supplies prompts, and Input Image supplies reference images.

## Frequently asked questions

**Which backend should I use for the Text → Image (High Quality) node?**

The Text → Image (High Quality) node supports multiple backends, each with different rendering strengths across photorealistic, illustrative, and stylized aesthetics. Run a few backends with the same prompt early in the project to compare. Lock in one backend per asset class and document the choice so the team uses it consistently.

**How does the Describe Image node differ from a captioning tool?**

The Describe Image node converts an image into structured or descriptive text, and its system prompt controls what is extracted: scene description, object list, spatial layout, material analysis, asset identification, style breakdown, or inferred function. The output is designed to feed back into prompt-driven image generation, not only to read as a human caption.

**Can I generate the same image twice from the same prompt?**

Yes. Fix the seed parameter to the same value and use the same backend to reproduce a result. Different seeds produce different variations. The Procedural Image Builder node is likewise deterministic: identical parameters always produce the same output unless the seed is changed.

**How do I prevent style drift across a batch of assets?**

Two patterns work. First, extract a description from a canonical reference with the Describe Image node, then feed that description into every batch prompt via Combine Text in [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes). Second, use Text + Image → Image in [Image Edit Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-edit-nodes) with a consistent style reference for every asset in the set.

**When should I use the Procedural Image Builder node instead of a text-to-image node?**

Use the Procedural Image Builder node when you need textures and materials built from algorithmic patterns such as Voronoi, Perlin noise, grids, and checkerboards, rather than AI synthesis. It suits tiling backgrounds, UI elements, and placeholder art, and its output width and height are configurable.

**Are generated textures from the Procedural Image Builder node tileable?**

The Tileable option ensures seamless wrapping. Enable it for textures intended for repeating UV maps, and leave it disabled for one-off images. Scale, seed, color palette, and pattern-specific controls adjust the visual result.

**What can I use a generated image for downstream?**

A polished image from the Text → Image (High Quality) node feeds downstream workflows including Image → 3D, Text + Image → Image, Edit Image with Text, and marketing assets. Descriptions produced by the Describe Image node can also supply metadata for Make3D or Auto Mesh Transform.


# Image Edit Nodes

Image editing nodes refine, restyle, and modify existing images using text instructions. Three core nodes range from fully generative edits to precise localized modifications, used heavily for iterati

## When to use image edit nodes

* **Refine generated images.** Take output from [Image Generation](/atlas-ai-studio-overview/node-index/image-nodes/image-generation-nodes) and adjust composition, style, or content with text instructions rather than re-rolling the entire generation.
* **Isolate subjects for 3D.** Use the generative edit nodes (Text + Image → Image) to remove backgrounds, isolate single objects, or clean up cluttered references before feeding into image-to-3D.
* **Apply style transfer.** Bring an existing image into a different aesthetic via Text + Image → Image with style-reference inputs.
* **Make surgical changes.** To modify a specific aspect (color, lighting, single element) without altering the rest of the image.
* **Generate variations from a reference.** Use Text + Image List → Image with multiple references to produce assets that combine aesthetic elements from several inputs.

Atlas provides powerful image-editing nodes that can accept **text + at least one image input**:

* **Text + Image → Image Node**
* **Text + Image List → Image Nodes**

These nodes share the same core idea:

* You provide an image (or multiple images) and a text instruction, and the node generates a modified image using the chosen backend.

They differ in how much freedom the model has:

* Text+ Image **→** Image and Text + Image List **→** Image nodes allow **fully generative edits**
* It can keep the **original image untouched except for the specified modification**.

## Nodes Overview

### Text + Image **→** Image <a href="#multimodal-node" id="multimodal-node"></a>

A flexible, image-conditioned generation engine.

* Requires text + image(s)
* Merges all input images with the instruction
* Allows scene transformations, object replacement, isolation
* Produces new images guided by the provided visuals
* Different backends = different image generation models

### Text + Image List **→** Image <a href="#multimodal-node" id="multimodal-node"></a>

A versatile, multi-reference image generation engine.

* Requires a text prompt + an array of up to 4 images
* Synthesizes multiple visual inputs based on complex text instructions
* Supports object isolation, composition of multiple scenes, targeted edits and perspective shifts
* Produces a single output image by referencing specific inputs (e.g., `image_0`, `image_1`)
* Can change aspect ratio to expand the scene
* Different backends (Gemini, Reve, Flux, etc.) offer varying levels of aesthetic quality and logic handling.

<figure><img src="/files/4x4MZxxlg0HIQmlro091" alt="" width="563"><figcaption></figcaption></figure>

***

## Inputs

#### Required: <a href="#required" id="required"></a>

* **Text prompt**
* **At least one image**

Optional:

* **Multiple images**&#x20;

These inputs can be manually uploaded or generated via other nodes.

## Backends (the AI model powering the node)

You may choose from multiple backend generative models. Each backend interprets images differently:

* Some are more realistic
* Some preserve structure better
* Some are more stylized
* Some adhere more strictly to prompts

Switching backend changes the artistic and structural behavior.

<figure><img src="/files/qsjMLnMvmQEeIpFaLK9T" alt="" width="563"><figcaption></figcaption></figure>

***

## Seed Control

Changing the seed gives new variations while keeping:

* the same structure
* the same instruction
* the same conditioning images

Keeping the seed = reproducible outputs.

***

## Example Use Cases:&#x20;

### Style Transformation of a Scene <a href="#a-style-transformation-of-a-scene" id="a-style-transformation-of-a-scene"></a>

Turn a normal bookshelf into a **fantasy enchanted bookshelf**.

<figure><img src="/files/Ow0i38EVrTGAZtcuEmVR" alt="" width="563"><figcaption></figcaption></figure>

### Variation with Seed Change <a href="#b-variation-with-seed-change" id="b-variation-with-seed-change"></a>

Same prompt + same image + same backends + **different seeds** = different variations.

<figure><img src="/files/P78jzAsrSR1o1ANCxhhQ" alt="" width="563"><figcaption></figcaption></figure>

### Restyling While Preserving Layout <a href="#c-restyling-while-preserving-layout" id="c-restyling-while-preserving-layout"></a>

Transform the bookshelf into an **old, damaged one** while keeping the structure.

<figure><img src="/files/cuX4qIMAHHbGX1fT82Li" alt="" width="563"><figcaption></figcaption></figure>

### Isolate an Object with Neutral Background <a href="#d-isolate-an-object-with-neutral-background" id="d-isolate-an-object-with-neutral-background"></a>

Perfect for Image → 3D workflows.

<figure><img src="/files/elYiQp5gA2dr9VvfHwgm" alt="" width="563"><figcaption></figcaption></figure>

### Multi-Image Conditioning for Object Replacement <a href="#e-multi-image-conditioning-for-object-replacement" id="e-multi-image-conditioning-for-object-replacement"></a>

Replace specific items on a shelf using three reference object images.

<figure><img src="/files/iXmj5IcO06mzCSusJckK" alt="" width="563"><figcaption></figcaption></figure>

#### Adding Details to a Generated Asset <a href="#a-adding-details-to-a-generated-asset" id="a-adding-details-to-a-generated-asset"></a>

Example: adding decorative elements or small props without changing the underlying concept.

<figure><img src="/files/0XsUmeExGDDSWOl7BzVQ" alt="" width="563"><figcaption></figcaption></figure>

#### Removing Objects <a href="#b-removing-objects" id="b-removing-objects"></a>

<figure><img src="/files/J7Ayft5eNO891IAOHUIx" alt="" width="563"><figcaption></figcaption></figure>

#### Changing the Design Style <a href="#c-changing-the-design-style" id="c-changing-the-design-style"></a>

You can restyle the asset into a new design language (e.g., Scandinavian / futuristic / gothic) while keeping the **same layout and camera**.

<figure><img src="/files/s4F7t6EVa3FlEwpsmzLL" alt="" width="563"><figcaption></figcaption></figure>

Changing the **seed** gives controlled alternatives of the *same style*.

<figure><img src="/files/GfM5042X0huUQkUnxAKd" alt="" width="563"><figcaption></figcaption></figure>

#### Changing Image Aspect Ratio <a href="#d-changing-image-aspect-ratio" id="d-changing-image-aspect-ratio"></a>

<figure><img src="/files/L5hO5ydvMdtdHF9BNX4h" alt="" width="563"><figcaption></figcaption></figure>

## Common pitfalls

* **Using a fully-generative node when you want a precise edit.** Text + Image → Image rewrites the whole image. If you only want to change a single element, use Edit Image with Text instead. Otherwise you'll get unintended changes elsewhere.
* **Conflicting text and image guidance.** When the text instruction contradicts the reference image, output quality drops. Make sure prompts align with the visual reference, especially when using multiple reference images.
* **Over-editing iteratively.** Each pass through an edit node loses some fidelity. After 3-4 chained edits, the image often degrades visibly. For complex changes, either combine instructions into a single edit pass or re-generate from a refined prompt.
* **Forgetting Text + Image List for multi-reference cases.** When combining style, subject, and lighting from different references, Text + Image List → Image accepts multiple images. Using Text + Image → Image with a single reference and trying to describe the others in text rarely works.

## Related nodes

* [Image Generation Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-generation-nodes) — the upstream source of images that often feed into edit nodes.
* [2D Post Processing Nodes](/atlas-ai-studio-overview/node-index/image-nodes/2d-post-processing-nodes) — finishing operations (camera framing, upscaling, background removal) after edits are complete.
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — image-to-3D nodes consume the output of edit operations. Edit nodes are commonly used to isolate subjects before 3D conversion.
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — Describe Image and Text Generation (LLM) can refine edit instructions; Create Image List supports multi-reference Text + Image List workflows.

## Frequently asked questions

**When should I use Text + Image → Image ?**

Text + Image → Image can be used for fully generative purposes: the output is a new image guided by your prompt and reference. Use it for major transformations (style transfer, composition changes, subject swaps). Also it can be used for fully conservative purposes: the original image is preserved except where the instruction explicitly changes things. Use it for surgical edits (color shift, single element change, minor cleanup).

**Can I use multiple reference images in a single edit?**

Yes, with Text + Image List → Image. The node accepts an image array (built via Create Image List in [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes)) and uses all references when generating output. Useful for blending style references with subject references.

**Can edit nodes upscale or change resolution?**

Edit nodes preserve resolution by default. For upscaling, use the dedicated upscale nodes in [2D Post Processing](/atlas-ai-studio-overview/node-index/image-nodes/2d-post-processing-nodes).


# 2D Post Processing Nodes

2D post-processing nodes in Atlas AI Studio for upscaling, outpainting, background removal, camera control, cropping, PBR material, and SVG output.

Post Process Nodes are used to refine, enhance, or extend generated images after the primary rendering step.

These tools help achieve better framing, higher resolution, clean cutouts, and extended compositions.

## When to use 2D post-processing nodes

* **Pre-3D cleanup.** Use Background Removal and Segmentation to isolate subjects before feeding [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) image-to-3D generators. Clean inputs produce dramatically better 3D output.
* **Upscaling for production resolution.** Generated images are often 1024×1024 or smaller. Use upscale nodes to bring outputs to print, video, or hero-asset resolutions.
* **Framing and composition fixes.** Use Camera Control to adjust perspective and Extend Image to outpaint missing parts of a composition without regenerating.
* **Texture finalization.** Use Multimodal and segmentation nodes to extract specific surface or material zones from a complex image, ready for [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) Apply Textures to Mesh.
* **Asset comparison and review.** Use Concatenate Images from [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) alongside camera-controlled variations for art-director review grids.

### Camera Control <a href="#camera-control" id="camera-control"></a>

The **Camera Control** node allows you to adjust the camera angle and framing of a generated image **without re-prompting**.

#### Adjustable Parameters <a href="#adjustable-parameters" id="adjustable-parameters"></a>

* **Rotate Left/Right:** `(-360 to 360)`
* **Move Forward:** `(0 to 10)`
* **Vertical Angle:** `(-1 to 1)`
* **Wide Angle Lens:** enable or disable
* **Seed:** generate different angle variations

<figure><img src="/files/mIifnSiRzoZYeZgiyglN" alt="" width="563"><figcaption></figcaption></figure>

This node is ideal for fine-tuning poses, correcting perspective, and improving shot composition.

### Extend Image <a href="#image-outpainting" id="image-outpainting"></a>

The **Extend Image** node expands an image beyond its original boundaries using Outpainting AI model backends.

* Input your image
* Provide a **prompt** describing how the new extended areas should look
* The model naturally blends extensions into the existing scene

Use this to:

* Widen the environment
* Add missing context
* Create cinematic aspect ratios
* Extend canvases for layout or UI needs

<figure><img src="/files/w3ZsUlba4dMvTGfv8MKC" alt="" width="563"><figcaption></figcaption></figure>

### Remove Image Background <a href="#remove-background" id="remove-background"></a>

Removes the background from an existing image and outputs two formats:

* **Binary PNG** (black/white mask)
* **Transparent PNG** (clean cutout)

Different **backends** produce different separation styles, so results may vary.\
Useful for:

* Isolating assets
* Preparing images for compositing
* Feeding a clean silhouette into other Atlas nodes

<figure><img src="/files/ekoF3gwTBlvoZzSu7ibr" alt="" width="563"><figcaption></figcaption></figure>

### Upscale Image <a href="#upscale-image" id="upscale-image"></a>

The **Upscale Image** node enhances resolution and detail.

* Input an image
* Set the **upscale factor**
* Choose from multiple backends for varying sharpness and texture behavior

Use this whenever you need:

* Higher fidelity output
* Cleaner textures
* Enhanced details for close-ups
* Large format renders or presentation images

<figure><img src="/files/Zd8NsNnN0Tzwy5n4VJAT" alt="" width="563"><figcaption></figcaption></figure>

### Find Images by Description

It allows you to query an image array using a text prompt and return only the images that match your description. This is especially useful when working with large sets of extracted images, such as those obtained from the **Extract Document Images** node.

You can specify the number of images you want to retrieve and choose the retrieval mode:

* **At Most N Retrieved** — returns up to the requested number of relevant images
* **Exactly N Retrieved** — forces the output to contain the exact number specified

<figure><img src="/files/yzymbkYfzz9KT4thKBaN" alt="" width="563"><figcaption></figcaption></figure>

This node is ideal when you need to filter a broad image collection for workflow-specific tasks.\
For example, you can provide an entire set of document images as input and use a prompt such as “images showing architectural elements” to retrieve only the relevant subset. The node then outputs the desired number of images that best match the prompt, allowing you to use them directly in multimodal nodes, reference-based generation, or any downstream image-processing steps within your workflow.

### Convert Image to SVG <a href="#image-to-svg" id="image-to-svg"></a>

The Convert **Image to SVG** node converts a raster image into a **vector-based SVG** output.

* Input: single image (e.g., logo, icon, black-and-white mask, simple graphic)
* Output: SVG file that can be scaled, edited, and reused in 2D or 3D workflows

For best results, start from images with **clear edges** and **high contrast** (e.g., black shapes on white background or simple flat-color drawings). The resulting SVG can then be refined in external vector tools or used directly in pipelines that expect curve or path-based input.

<figure><img src="/files/QBp72B1QMliB0nplZLTF" alt="" width="563"><figcaption></figcaption></figure>

### Split Image into Layers

Decomposes a single image into multiple RGBA layers, such as separating the foreground from the background.

<figure><img src="/files/FLXKBTCFIimMebTox36o" alt="" width="563"><figcaption></figcaption></figure>

### Split Alpha

Separates the transparency (alpha) channel from an image, providing the color data and the mask as two separate outputs.

<figure><img src="/files/htBlULryBLuuhjtBrRPj" alt="" width="563"><figcaption></figcaption></figure>

### Combine Alpha

Merges a color image with a separate grayscale image to create a single image with transparency.

<figure><img src="/files/NMGWArsEXupfjTebQ3gC" alt="" width="563"><figcaption></figcaption></figure>

### Split Image Channels

Breaks an image down into four individual grayscale images representing its Red, Green, Blue, and Alpha channels.

<figure><img src="/files/43nkZjC2rxPkPDvKGXnx" alt="" width="563"><figcaption></figcaption></figure>

### Combine Image Channels

Takes up to four separate grayscale images and combines them into a single RGBA color image.

<figure><img src="/files/qH8329DsB98YVsrrvn8j" alt="" width="563"><figcaption></figcaption></figure>

### Threshold Binary Mask

Converts an image to grayscale and applies a threshold to create a high-contrast black-and-white binary mask.

<figure><img src="/files/YJdIFj5R2izdxoQeUd8L" alt="" width="563"><figcaption></figcaption></figure>

### Crop Image Manual

Allows you to crop an image by manually specifying the left, top, right, and bottom coordinates.

<figure><img src="/files/YX9YDALQZOLTbBnHy46w" alt="" width="563"><figcaption></figcaption></figure>

### Crop Image Auto

Automatically crops an image based on its transparency or a provided mask, removing unnecessary empty space.

<figure><img src="/files/zaX3vpdGdgyFFQNRaWzk" alt="" width="563"><figcaption></figcaption></figure>

### Simple Image Resize

Changes an image's dimensions to a specific width and height with options to preserve the aspect ratio.

<figure><img src="/files/unHh70A7y64ufE6GSju7" alt="" width="563"><figcaption></figcaption></figure>

### Depth Estimation

Analyzes an image to create a depth map, where the brightness or color represents the distance of objects from the camera.

<figure><img src="/files/2FbB1Cztx9ozaJpBpSGZ" alt="" width="563"><figcaption></figcaption></figure>

* Provides depth map output in EXR format for download.

### Fast Normal from Depth

A quick algorithm that generates a normal map (used for lighting and 3D effects) directly from a grayscale depth map.

<figure><img src="/files/dxNq3RSyQZLegrMK6qk6" alt="" width="563"><figcaption></figcaption></figure>

### Image Filters

Applies algorithmic image transformations without requiring a generative backend. The node provides 69 filters organized into eight categories, each with built-in parameter guidance for rapid adjustment.

<figure><img src="/files/Ti4kj8VJwL5W2ZLIVy6k" alt="" width="563"><figcaption></figcaption></figure>

* **Input**: Single image (any resolution)
* **Output**: Filtered image matching input dimensions
* **Filter categories**: Noise and Grain, Blur and Smooth, Color/Tone, Sharpen, Distortion, Degrade/Stylize, Weather, Blend Modes
* **Selection interface**: Single dropdown menu listing all 69 filters
* **Parameters**: Context-sensitive sliders and controls appear automatically based on the selected filter
* **Deterministic**: Produces identical output for identical inputs and settings; no randomness or inference involved
* **Performance**: Processes locally; suitable for real-time preview and iteration

Useful for: Rapid texture variation, color grading screenshots for marketing, adding atmospheric effects to concept art, prototyping visual effects before implementing them in-engine, and batch-processing asset alternatives without waiting for generative models.

**Example:** Examples of different Image Filters applied to the same input across multiple filter categories.

<figure><img src="/files/CaPmM8f4tHVXxcGZ0leA" alt="" width="563"><figcaption></figcaption></figure>

### Material Generation

Generates physically based rendering (PBR) material maps from text prompts, reference images, or existing textures. The node outputs separate map channels (base color, metallic, roughness, normal) compatible with standard game engines and 3D tools.

<figure><img src="/files/ICgIKvtFG81gKmD1H8oB" alt="" width="563"><figcaption></figcaption></figure>

* **Input**: Text prompt, reference image, or existing texture depending on selected mode
* **Output**: Four separate PBR map outputs—base color, metallic, roughness, and normal—each connectable independently
* **Modes**: Generate (prompt-to-material), Predict (image-to-material), Extract (texture-to-channels)
* **Map resolution**: Configurable output size for each material channel
* **Direct export**: Maps route directly to downstream nodes or export for use in Blender, Unreal Engine, Unity, and other PBR-compatible tools
* **Channel isolation**: Individual map outputs allow selective regeneration or replacement of specific channels without affecting others

Useful for: Rapid material prototyping during blockout phases, generating tileable surface variants from written descriptions, extracting PBR channels from photographs or scanned textures, and creating material libraries for environmental art without manual authoring.

**Example:** extract PBR maps from a reference texture, or describe a material (e.g. 'magical cracks') to generate base color, metallic, roughness, and normal maps.

<figure><img src="/files/gJnB31rDc0Bis5KMWhBo" alt="" width="563"><figcaption></figcaption></figure>

### Text to SVG Node

The **Text to SVG** node generates scalable vector graphics (SVG) directly from a text prompt, producing resolution-independent assets suitable for UI elements, icons, and game graphics.

* Input a **text prompt** describing the desired vector graphic
* Outputs an **SVG file** that can be scaled infinitely without quality loss
* Useful for generating logos, icons, symbols, and geometric assets
* Vector format integrates cleanly into game engines and design tools
* Ideal for rapid prototyping of UI elements before final art production

<figure><img src="/files/1MMIuSUz1nZiEXCCNT9a" alt="" width="563"><figcaption></figcaption></figure>

**Useful for:** generating placeholder icons, creating scalable HUD elements, prototyping menu graphics, and producing clean vector assets for 2D games.

**Example:** Text to SVG node examples showing output variation across different backends.

<figure><img src="/files/YeRdEa1r4uRFLRBSUirA" alt="" width="563"><figcaption></figcaption></figure>

### Normal Map Convention Convert Node

The **Normal Map Convention Convert** node flips the Y-axis channel of a normal map to switch between DirectX (Y-down) and OpenGL/glTF (Y-up) conventions. Different game engines and 3D tools use opposing standards for the Y-axis in tangent-space normal maps; this node reconciles those differences without external software.

* **Input:** a normal map texture
* **Output:** the same normal map with Y-axis inverted
* Automatically detects or allows manual selection of source convention
* No quality loss—only inverts the green channel
* Essential when importing assets between engines with different normal map standards
* Prevents lighting artifacts caused by reversed surface normals

<figure><img src="/files/moVeymTgzlzDPZaSV6qn" alt="" width="563"><figcaption></figcaption></figure>

**Useful for:** preparing AI-generated or imported normal maps for Unity, Unreal Engine, Godot, or other engines with differing tangent-space conventions.

### Smart Resize Image

The **Smart Resize Image** node intelligently reframes and crops images to a target size using AI-guided composition. Unlike standard cropping, it interprets a text prompt to decide which parts of the image to preserve or emphasize.

* **Input:** source image and target dimensions
* **Prompt:** describe the subject or area to focus on (e.g., "character's face", "weapon", "background landscape")
* **Output:** resized image with content-aware framing
* Preserves important visual elements based on semantic understanding
* Useful for generating portrait thumbnails, UI icons, or adaptive aspect ratios from wide screenshots

<figure><img src="/files/n7dHaqexpzcWOdhLbLSj" alt="" width="563"><figcaption></figcaption></figure>

Ideal for creating multiple aspect ratios from a single asset, focusing character portraits, or adapting cutscene frames for different screen formats.

### Paste Crop Into Image

The **Paste Crop Into Image** node composites an edited crop back into its source image, using alpha-aware blending and automatic resizing to fit the target region.

* **Source and crop inputs.** Takes the original image and the modified crop to be reinserted.
* **Region mask input.** Accepts a PNG region mask (white rectangle on black) defining where the crop is placed, matching the mask produced for Crop Image Auto.
* **Alpha-aware compositing.** Respects transparency in the crop so edited regions blend cleanly against the source.
* **Automatic resizing.** Scales the crop to match the target region when dimensions differ.
* **Out-of-bounds clipping.** Clips portions of the crop that extend past the canvas edge.
* **Output.** Returns the recomposited full image.

Useful for: reinserting locally edited details—repainted armor, altered NPC faces, adjusted signage—back into a larger scene without regenerating the whole asset.

**Example:** Paste Crop Into Image example: edit only the selected object and paste it back into the original image, enabling an inpainting-style workflow.

<figure><img src="/files/aIp0Dpb76qdpA8N8pGi5" alt="" width="563"><figcaption></figcaption></figure>

### Ellipse Mask

The **Ellipse Mask** node defines an elliptical editable region on a reference image, producing a mask that constrains where downstream editing nodes apply changes.

<figure><img src="/files/O0dQdAms6jsiAF8Dggaq" alt="" width="563"><figcaption></figcaption></figure>

* **Input.** A reference image on which the elliptical region is placed.
* **Output.** A mask isolating the elliptical area for use by inpainting, region-edit, or compositing nodes.
* **Visual mask picker.** Drag and resize an overlay directly on the reference image to position the ellipse.
* **Live coordinate inputs.** Adjust center, width, and height numerically alongside the interactive overlay for precise placement.
* **Pairs with Rect Mask.** Use Ellipse Mask for rounded or organic regions and Rect Mask for straight-edged selections.

Useful for: masking faces, shields, planets, spell effects, or other circular elements for targeted edits and asset variation without regenerating the full image.

### Rect Mask

The **Rect Mask** node defines a rectangular editable region on a reference image, restricting downstream edits to the masked area while leaving the rest of the image unchanged.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FubqKzNVpbaUEFyGTmuaZ%2Frect_mask_video.mp4?alt=media&token=3c27fb95-0855-433c-9b9e-adbcdf97d027>" %}

<figure><img src="/files/kLEzbT0DPJ4QKGkCkih5" alt="" width="563"><figcaption></figcaption></figure>

* **Input.** A reference image to mask.
* **Output.** A mask defining the selected rectangular region, passed to downstream editing nodes.
* **Visual mask picker.** Drag and resize an overlay directly on the reference image to set the region interactively.
* **Live coordinate inputs.** Numeric fields update in sync with the overlay for precise placement and sizing.
* **Localized edits.** Confines regeneration or replacement to the masked bounds, preserving surrounding detail.

Useful for: swapping in-scene props such as replacing a painting on a wall, retexturing a specific surface, or generating asset variations within a fixed area of an environment.

### Custom Image Format Output

The **Custom Image Format Output** node exports images to a chosen file format with per-format encoding controls, giving fine-grained management over quality and file characteristics for pipeline delivery.

<figure><img src="/files/4Xo50cKMCLLL85aGvWTx" alt="" width="563"><figcaption></figcaption></figure>

* **Supported formats.** Exports as JPEG, TGA, WebP, TIFF, or AVIF, each with its own applicable settings.
* **Quality control.** Adjustable quality and subsampling for lossy formats to balance fidelity against file size.
* **Compression settings.** Per-format compression options for formats that support them.
* **Alpha handling.** Configurable transparency behavior for formats with alpha channels.
* **Lossy source warning.** Alerts when exporting from an already-lossy source, flagging potential compounding quality loss.

Useful for: delivering game-ready textures, UI art, and sprite assets in the exact format and compression profile a target engine or platform requires.

### Invert Mask

The **Invert Mask** node flips a grayscale or binary mask at the pixel level, swapping editable and preserved regions. This isolates the opposite of a segmentation or background-removal result without re-running prompts or hand-painting inverted shapes.

<figure><img src="/files/hWqBYhR6lmLR0C6zwD1P" alt="" width="563"><figcaption></figcaption></figure>

* **Input.** A grayscale or binary mask, typically produced by segmentation or background-removal nodes.
* **Output.** The inverted mask, with editable and preserved regions swapped.
* **Pixel-level operation.** Flips mask values directly, preserving resolution and edge detail from the source mask.
* **No re-prompting.** Reuses an existing mask rather than regenerating a new one for the opposite region.
* **Typical use.** Convert a "subject" mask into a "background" mask (or vice versa) to target the complementary area in downstream compositing or inpainting.

Useful for: isolating backgrounds behind isolated characters, masking surrounding scenery for environment edits, and preparing complementary regions for asset variation.

## Common pitfalls

* **Upscaling before refining.** The Upscale Image node locks in artifacts and noise present in the source image. Complete edit and cleanup passes first, then run Upscale Image at the end of the workflow.
* **Outpainting too aggressively.** The Extend Image node works best for modest extensions. Asking it to outpaint 3-4x the original area produces visible seams and style drift. For large extensions, run several smaller passes or regenerate at the larger target size.
* **Background removal on complex compositions.** The Remove Image Background node produces ragged cutouts when a highly detailed background blends with the subject. When the subject-background boundary is ambiguous, isolate the subject with a generative edit instead.
* **Feeding a flat view into image-to-3D.** A head-on generated image gives image-to-3D conversion no perspective to work from. Run it through the Camera Control node to add a three-quarter view before passing it to the Mesh Nodes.
* **Resizing without composition awareness.** The Simple Image Resize node changes dimensions to a fixed width and height and can crop away important subjects. When the target aspect ratio differs significantly from the source, use the Smart Resize Image node with a prompt describing the subject to preserve.
* **Vectorizing low-contrast images.** The Convert Image to SVG node produces noisy paths from soft or photographic input. Feed it images with clear edges and high contrast, such as flat-color graphics or black-and-white masks.

## Related nodes

* [Image Generation Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-generation-nodes) — the upstream source of images that often need post-processing.
* [Image Edit Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-edit-nodes) — generative-edit alternatives for cases where post-processing nodes are not enough, such as isolating a subject when the background boundary is ambiguous.
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — image-to-3D nodes that consume post-processed images. Image quality at this stage materially affects 3D output, and the Material Generation node produces PBR maps for the same pipeline.
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — Concatenate Images, Break Images Array, and Create Image List for batching and comparison workflows.

## Frequently asked questions

**What is the maximum upscale factor?**

The maximum upscale factor of the Upscale Image node depends on the selected backend. For large increases, chaining two passes can produce smoother results than a single high-factor pass.

**Can the Remove Image Background node handle hair, fur, or fine detail?**

It handles hair and fur with some softness on edges. Because different backends produce different separation styles, results vary. For pixel-perfect cutouts on hero assets, do a manual cleanup pass in an external image editor after the automated removal.

**How do I outpaint an image to a new aspect ratio?**

Use the Extend Image node with a prompt describing how the extended areas should look. For dramatic shifts, such as square to widescreen, a guiding prompt keeps the extended area stylistically coherent, and several smaller passes reduce seams.

**Does the Camera Control node work on real photographs?**

Yes. It adjusts angle and framing of any 2D image, including photographs, without re-prompting. Extreme angle changes may introduce artifacts, so modest adjustments are most reliable.

**How do I generate PBR material maps from a texture or prompt?**

Use the Material Generation node. Generate mode takes a text prompt, Predict mode takes a reference image, and Extract mode takes an existing texture. Base color, metallic, roughness, and normal are separate outputs, so you can regenerate or replace one channel without affecting the others.

**Which SVG node should I use?**

The Text to SVG Node generates a vector graphic from a text prompt. The Convert Image to SVG node converts a raster image, such as a logo or black-and-white mask, into vector paths and needs clear edges and high contrast. Both produce assets that scale without quality loss.

**Can I crop a region, edit it, and paste it back?**

Yes. Use the Crop Image Auto node to crop based on transparency or a mask, edit the crop, then use the Paste Crop Into Image node to composite it back. The region mask is a white rectangle on black matching the Crop Image Auto output; the crop is alpha-blended and auto-resized to the target region, and portions past the canvas edge are clipped. Use the Crop Image Manual node to specify left, top, right, and bottom coordinates directly.

**Do the Image Filters run without a generative backend?**

Yes. The Image Filters node applies algorithmic transformations locally and is deterministic, producing identical output for identical inputs and settings. This makes it suitable for real-time preview and batch iteration without inference.


# Mesh Nodes

Mesh nodes in Atlas AI Studio generate, transform, texture, optimize, rig, and animate 3D meshes from images and multi-view references.

Mesh Nodes operate on 3D geometry created within Atlas workflows.

They handle **mesh generation**, **transformation**, **cleanup**, **optimization**, and **scene assembly**, ensuring that assets become production-ready and correctly scaled.

## When to use mesh nodes

* **Generate 3D from 2D.** Use Image → 3D or Multi-View → 3D as the entry point for new 3D content from concept art or multi-view character sheets. See [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices) for the deep dive on backends, pitfalls, and post-processing.
* **Make AI-generated meshes game-ready.** Use Optimize Mesh, Reduce Polycount, and Auto Transform Mesh to bring dense AI-generated geometry into game-engine territory (polycount, topology, scale, pivot).
* **Re-texture an existing asset.** Use Re-texture Mesh with a style reference image to restyle a model without regenerating geometry. Useful for variant generation.
* **Bake high-poly detail onto low-poly.** Generate at high quality, retopologize, bake normal and PBR maps onto the low-poly mesh. The Bake High-Poly to Low-Poly workflow covers this end to end.
* **Rig and animate generated characters.** Use Rig Humanoid Mesh + Animate Rigged Model for full character pipelines from concept art to game-ready animated NPCs.
* **Assemble scenes from generated meshes.** Use Compose 3D Scene to arrange multiple generated meshes into a layout based on a text prompt.

### Image **→** 3D <a href="#image-to-3d" id="image-to-3d"></a>

The **Image** **→ 3D** node converts a single reference image into a **3D mesh**.

* Takes an **image input** and reconstructs a 3D model.
* Multiple **backends** are available (low-poly, high-detail, quad-based, triangle-based, stylized, realistic).
* Output quality varies based on backend.
* Use this node whenever you want to turn a concept art, render, or isolated product image into a mesh.

More details are provided in the **Image**->**3D** sub-section.

<figure><img src="/files/TYw3NjHgVmeewa6MLRKo" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/YIlJJndtkxWM7aLbdqMq" alt="" width="563"><figcaption></figcaption></figure>

### Fast Image **→** 3D

Rapidly generates a 3D reconstruction from a single 2D image in approximately 30 seconds.

* Prioritizes speed and rough volume, making it perfect for quick concept prototyping.
* Features options for PBR (Physically Based Rendering) support to include lighting information in the output.
* Backends: SAM3D or Hunyuan3D Rapid.
* Ideal for: Creating background assets or quickly visualizing 2D concepts in 3D space.

<figure><img src="/files/2qe0MLQgGx1HSgUpUm4Y" alt="" width="563"><figcaption></figcaption></figure>

### Image **→** 3D (With Fallback)

A production-oriented 3D generation node that attempts reconstruction using multiple AI backends sequentially.

* If the primary backend fails, the node automatically retries with secondary "fallback" models to ensure a result is always produced.
* Offers deep customization for quality, face limits, and PBR textures across up to 4 different backend attempts.
* Inputs: Input Image, Multiple Backend choices, Retry Count.
* Ideal for: Production pipelines where reliability and successful generation are critical.
* Ideal for UGC use cases where reliability is critical

<figure><img src="/files/5fkcwk6ksIWJ42K0Kf6j" alt="" width="563"><figcaption></figcaption></figure>

### Multi-View **→** 3D

Reconstructs a 3D mesh from a set of consistent images showing an object from different standardized angles.

* Provides significantly higher geometric accuracy and consistency than single-image generation.
* Accepts specific views (Front, Right, Left, Back, Top, Bottom) to build a comprehensive 3D volume.
* Backends: Includes industry leaders like Tripo v3.0 and Meshy v6.
* Ideal for: Creating high-fidelity 3D models from character sheets or multi-angle photographs.

<figure><img src="/files/sXdyUfVfO7cQVgx3POu1" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/GPFrMX5yoEY5H49xsjBN" alt="" width="563"><figcaption></figcaption></figure>

### Retexture Mesh <a href="#texture-mesh" id="texture-mesh"></a>

The Re-t**exture Mesh** node regenerates or restyles the texture of an existing mesh.

* Input:
  * A **reference style image**
  * An **existing mesh**
* The resulting mesh receives a texture that matches the reference image.
* Ideal for creating **style variations** or re-texturing an asset.

Example: Change the style of the asset by using a new 2D concept as the texture reference.

<figure><img src="/files/Tmj3itGftcrZBqZAdC4J" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/fr1YPbOnd3zGG0GIMRHq" alt="" width="563"><figcaption></figcaption></figure>

* Supports multiple style images as input (front, left, right, and back views) for improved retexturing accuracy and directional control.

<figure><img src="/files/ico96pBxJtDEkmZO6Nvz" alt="" width="563"><figcaption></figcaption></figure>

* Additional generation arguments are available with some backends for fine-tuning retexturing behavior.

### Extract Texture Maps <a href="#auto-mesh-transform" id="auto-mesh-transform"></a>

Separates the visual components of a GLB mesh into individual PBR (Physically Based Rendering) texture files.

* Extracts the Base Color, Roughness, Metallic, and Normal maps as individual 2D images.
* Allows you to isolate and edit specific material properties (like making a surface shinier or adding bump detail) in 2D.
* Outputs: Base Color Image, Roughness Image, Metallic Image, Normal Map Image.

<figure><img src="/files/E2yf7IUoGtdipLXjSv8f" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/BqqvUDwahaHjO8rNh9l9" alt="" width="563"><figcaption></figcaption></figure>

### Apply Textures to Mesh <a href="#auto-mesh-transform" id="auto-mesh-transform"></a>

Replaces or assigns 2D texture images to the specific material slots of a 3D model.

* Takes a geometry-only or existing mesh and applies provided images to its UV-mapped surfaces.
* Allows for selective replacement; only the maps you provide (e.g., just the Base Color or just the Normal Map) will be updated.
* Inputs: Input Mesh, Base Color, Roughness, Metallic, Normal Map.
* Ideal for: Re-assembling a model after editing its texture maps in 2D.

<figure><img src="/files/P3zuFcSCxXhr52canjOe" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/TDGo8VlupEooXIc9nA70" alt="" width="563"><figcaption></figcaption></figure>

* Accepts a normal map convention setting and automatically converts DirectX maps to the glTF/OpenGL convention when needed.

### Auto Transform Mesh

The **Auto Transform Mesh** node is one of the most essential post-processing tools.

It automatically:

* Applies **semantic world-scale correction**
* Adjusts **dimensions** based on object type
* Repositions the **origin**
* Prepares the asset for proper placement in a game engine, CAD tool, or Atlas pipeline

Use this node immediately after generating a model.

<figure><img src="/files/GVaH8Kzww1dcfK0IzX5Y" alt="" width="563"><figcaption></figcaption></figure>

### Mesh Multi-View Render

Captures standardized 2D snapshots of a 3D model from multiple specified camera angles.

* Uses yaw (horizontal) and pitch (vertical) coordinates to define exact rendering perspectives.
* Essential for creating character sheets or preparing views for texture re-projection workflows.
* Outputs: An array of images and the corresponding camera matrices used for the render.

Note: This is a specialized tool typically used in advanced multi-view pipelines.

<figure><img src="/files/zhjcSZRxoGOo8ZBctvnP" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/fHg0bYz5e3Cti6Cj1WWp" alt="" width="563"><figcaption></figcaption></figure>

### Create Occlusions Mask

Generates UV-baked visibility maps that identify which surface areas are visible from specific camera views.

* Detects areas of the mesh that are hidden or shadowed, preventing textures from "smearing" onto unseen geometry.
* Crucial for high-quality texture projection where multiple 2D views must be blended seamlessly.
* Inputs: Input Mesh, Camera Matrices (from the Multi-View Render node).
* Outputs: An array of grayscale occlusion masks.

<figure><img src="/files/zL0qC1YiaBOhHdvqPVLd" alt="" width="563"><figcaption></figcaption></figure>

### Project Multi-View Images to Mesh

Transfers color data from a set of 2D images back onto the 3D surface of a mesh using UV coordinates.

* Blends multiple perspectives together while using occlusion masks to ensure textures are only applied to visible surfaces.
* Includes "Softmax Sharpening" to control the clarity and transition between different projected views.
* Inputs: Mesh, Image Array, Camera Matrices, Occlusion Masks.

Ideal for: Texturing a 3D model using AI-generated or edited 2D reference images.

<figure><img src="/files/rhguucikmcAfD4qM7oEx" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/bZgXbbBQBq0FdR0m6VyY" alt="" width="563"><figcaption></figcaption></figure>

### Optimize Mesh <a href="#remesh-model" id="remesh-model"></a>

Reconstructs the mesh topology and produces a cleaner, more manageable geometry structure.

Reduces or increases polygon count while maintaining the model’s overall shape.

* Topology options:
  * **Triangle** for most general uses.
  * **Quad** for modeling workflows.
* Input: target polygon count

This node is ideal for preparing AI-generated meshes that come in overly dense or irregular form.

<figure><img src="/files/WPbSYPquY5F4NVfIpapK" alt="" width="563"><figcaption></figcaption></figure>

* Some backends offer voxel-based remeshing with voxel size, narrow band width, and surface projection controls as an alternative to target polygon count and topology selection.

<figure><img src="/files/SUVBjuSThV2jaACyLSzM" alt="" width="563"><figcaption></figcaption></figure>

### Mesh BBox Fit <a href="#mesh-bbox-fit" id="mesh-bbox-fit"></a>

Scales a mesh **non-uniformly** so that its **world-space bounding box** exactly matches specified X, Y, Z dimensions.

* Input: target width, height, depth
* Output: scaled mesh with exact real-world bounding box

Useful for making an asset match required dimensions precisely.

<figure><img src="/files/yHbtSOT3nQ6WrhDqSyWl" alt="" width="563"><figcaption></figcaption></figure>

### Set Mesh Origin

Sets the mesh origin based on bounding-box parameters.

* Input: desired origin position (e.g., upper bound, bbox center)
* Output: repositioned mesh origin used for clean pivoting and placement

This is important for alignment, snapping, and scene assembly.

<figure><img src="/files/gAzg0GenPlDDo24ek3Gs" alt="" width="563"><figcaption></figcaption></figure>

### Text to Origin

Uses a **text instruction** to modify a mesh's origin.

Examples:

* “Set origin to bottom center”
* “Move pivot to the front face”
* “Place origin at the geometric center”

Helpful when precise manual origin adjustments are needed.

<figure><img src="/files/gPAgPyxY48luXqMT80E0" alt="" width="563"><figcaption></figcaption></figure>

### Rotate Mesh Towards Axis

Rotates the mesh around **(0, 0, 0)** so that it faces a selected axis.

* Supports presets such as:
  * Face −Y
  * Face +Y
  * Face −X / +X
  * Face −Z / +Z
* Assumes the input mesh initially faces **−Y**.

Ensures consistent orientation across generated assets.

<figure><img src="/files/a1G4SW2IKjojBqMlYsWy" alt="" width="563"><figcaption></figcaption></figure>

### Compose 3D Scene

Takes your generated meshes and arranges them into a **simple 3D scene** based on a text prompt.

* Input meshes + prompt
* Output: arranged layout with placement, rotation, spacing
* Useful for quick scene design or previews

<figure><img src="/files/WQSs5xNJZycjDZtcKUoe" alt=""><figcaption></figcaption></figure>

<figure><img src="/files/itlRinzuDG1FKXcRIJcc" alt="" width="563"><figcaption></figcaption></figure>

### Mask to Spline

Converts a **binary mask** into a spline.

* Requires input binary mask where your desired object is **white**.
* Outputs a spline based on the **largest connected white region**.
* Mask can be prepared via **Text+Image -> Image nodes** beforehand.

Used for shape extraction, outline-based modeling, or path generation.

<figure><img src="/files/7tONI8DKMhYOj3QOSxtm" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/aeZbzpTtOg8ndcrzL9cB" alt="" width="563"><figcaption></figcaption></figure>

### Seperate Object Parts

Automatically segment a 3D mesh into individual parts based on geometric and semantic analysis.

* Uses AI to identify and detach discrete components (e.g., separating a character's clothing or a machine's parts) into a segmented GLB.
* The primary backend supports up to 30k faces, while a fallback handles denser meshes up to 1.5M faces.
* Inputs: Input Mesh, Backend Selection, Seed.
* Ideal for: Modifying or isolating specific pieces of a combined AI-generated model.

<figure><img src="/files/UjkOwx38MHy3F4hf2Of6" alt="" width="563"><figcaption></figcaption></figure>

### Rig Humanoid Mesh

Automatically generates a skeletal structure and skin weights for humanoid character models.

* Creates an armature that allows the character to be posed and animated.
* Requires the model to be in a standard pose (T-pose or A-pose) and calibrated for height in meters for accurate skeletal placement.
* Outputs: Rigged Mesh, plus preset "Walking" and "Running" previews for instant testing.

<figure><img src="/files/hXTAHBt3yHySuvxKYfdY" alt="" width="563"><figcaption></figcaption></figure>

### Animate Rigged Model

Applies motion data and specific animations to a previously rigged character mesh.

* Utilizes a massive library of preset actions ranging from daily movements to combat and dancing.
* Requires metadata from a compatible rigging node to correctly map animations to the skeleton.
* Inputs: Rigged Model Metadata, Animation Selection (e.g., Fighting, Dancing, WalkAndRun).
* Ideal for: Quickly bringing static characters to life for games or cinematics.

<figure><img src="/files/39EQ8NyqdQQrZl9v2YnC" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/fwpyWdut1c7FUSb5dmRl" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/c77VxP0O8vAyAuQmkfy4" alt="" width="563"><figcaption></figcaption></figure>

### Omnipart

Segments an image and generates **separate mesh parts**.

* Outputs:
  * Segmentation map
  * Individual GLBs for each part
  * Optional merged mesh
* Ideal for breaking an asset into modules or components.

Perfect for kitbashing, modular asset workflows, or creating parametric components from a single image.

<figure><img src="/files/dbCNHki6wmN2XfPTMa79" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/6vXpy4lbDpbzWQ66GMLY" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/qK7AjJQ4trC6NPhAmPIv" alt="" width="563"><figcaption></figcaption></figure>

### Bake High-Poly to Low-Poly&#x20;

Transfers detail from a high-polygon mesh to a low-polygon proxy by baking normal maps and PBR texture channels. The node outputs standard texture maps that can be applied to game-ready geometry for efficient real-time rendering.

<figure><img src="/files/Gz1wuqouh4kt48RXosud" alt="" width="392"><figcaption></figcaption></figure>

* **High-poly mesh input** — source geometry containing sculpted or subdivided detail
* **Low-poly mesh input** — target game mesh with simplified topology and UV layout
* **Texture resolution** — explicit pixel dimensions for baked maps; no longer auto-detected from source
* **Bake margin** — scales automatically with resolution to prevent seam artifacts at higher output sizes
* **Output channels** — normal, albedo, roughness, metallic, and ambient occlusion maps
* **UV requirement** — low-poly mesh must have non-overlapping UVs in the 0–1 space

Useful for generating optimized hero assets, LOD transitions, and production-ready character models from sculpted prototypes.

* **AO samples** — adjustable sampling count for ambient occlusion baking; higher values improve quality at the cost of longer bake time
* **Texture margin** — manual control over edge dilation; Auto mode scales margin with resolution

**Example:**&#x20;

<figure><img src="/files/SigDO481eQn15RuWqaza" alt="" width="563"><figcaption></figcaption></figure>

* Inputs:
  * **High-poly mesh** with detailed texture
  * **Low-poly mesh** (typically from Optimize Mesh)
* Uses **Mesh BBox Fit** to align the scale and position of both meshes before baking.
* The **Bake High-Poly to Low-Poly** node projects surface detail from the high-poly texture onto the low-poly UV layout.
* Output: a low-poly mesh with baked texture that retains the appearance of the original high-resolution asset.
* Typical workflow: generate or import dense mesh → optimize geometry → align bounding boxes → bake texture detail.

Useful for creating game-ready assets from AI-generated or sculpted high-detail models, maintaining visual quality while meeting real-time performance budgets.

### Mesh → Maya USDZ

The **Mesh → Maya USDZ** node exports a mesh to a Maya-compatible **USDZ** file for downstream DCC work.

* Takes a **mesh input** and writes a **USDZ file** as output.
* Preserves material channels including **base color, ORM, normal, and UV** data.
* Produces files structured for compatibility with Maya-based pipelines.
* Supports USDZ download for handoff between the studio and external tooling.
* Pairs with the **USDZ → Mesh** node for round-trip exchange of geometry and materials.

Useful for: moving AI-generated or optimized game meshes into a Maya workflow for further authoring, cleanup, or integration into an existing asset pipeline.

<figure><img src="/files/QPpt1AHM0xu3wgbsccg0" alt="" width="563"><figcaption></figcaption></figure>

**Example:** Example use case for the Mesh → Maya USDZ node.

<figure><img src="/files/LbiYBN62bW7GQO13q0L3" alt="" width="563"><figcaption></figcaption></figure>

### USDZ → Mesh

The **USDZ → Mesh** node imports a USDZ file as a mesh for use in downstream nodes.

* Takes a **USDZ file input** via upload and converts it into an editable mesh.
* Preserves material data on import, including **base color, ORM, and normal** channels.
* Retains **UV** coordinates from the source file.
* Outputs a mesh compatible with retexturing, optimization, and scene composition nodes.
* Pairs with **Mesh → Maya USDZ** for round-trip exchange of assets.

Useful for: bringing externally authored or DCC-exported USDZ assets into a workflow for editing, variant generation, or preparing game-ready geometry.

<figure><img src="/files/yyrftEm2BUj3MXkAWqdb" alt="" width="563"><figcaption></figcaption></figure>

## Common pitfalls

* **Skipping Auto Transform Mesh.** Generated meshes often have non-standard pivot, scale, and orientation. Without normalization they are hard to use in procedural assembly or engine integration. Run Auto Transform Mesh after generation to apply semantic world-scale correction, adjust dimensions by object type, and reposition the origin.
* **Using high-detail backends for game targets.** High-detail backends produce dense meshes that need retopology and baking before they are game-ready. For real-time targets, start with low-poly or stylized backends and skip the bake step.
* **Optimizing before texturing.** Polycount reduction can damage UV layouts. Either texture first then optimize with re-baking, or use the Bake High-Poly to Low-Poly workflow, which aligns meshes with Mesh BBox Fit and projects high-poly detail onto the low-poly UV layout in the correct order.
* **Generating from poor reference images.** 3D generation quality tracks input image quality. Process reference images through [2D Post Processing](/atlas-ai-studio-overview/node-index/image-nodes/2d-post-processing-nodes) (background removal, perspective adjustment, isolation) before feeding mesh nodes. This applies to any image-to-video AI node or concept-art-to-mesh workflow.
* **Forgetting Image → 3D (With Fallback) for production UGC.** Single-backend generation can fail on unusual inputs. For player-facing UGC where reliability matters more than per-asset quality, use Image → 3D (With Fallback), which retries reconstruction across multiple backends automatically to ensure a result is produced.
* **Rigging non-standard poses.** Rig Humanoid Mesh requires the model to be in a standard T-pose or A-pose and calibrated for height in meters for accurate skeletal placement. Feeding an arbitrary pose produces unreliable armatures and skin weights.
* **Projecting textures without occlusion masks.** When texturing with Project Multi-View Images to Mesh, omitting the masks from Create Occlusions Mask lets color data smear onto surfaces hidden from a given camera view. Generate occlusion masks first so texture data is only applied to visible geometry.

## Related nodes

* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — Input Image, Input Images (for multi-view), and Input Mesh feed mesh-node workflows.
* [Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes) — generate and refine 2D references before feeding mesh nodes. Image quality determines 3D quality.
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — Split Character Sheet and Create Image List prepare multi-view inputs; LLM nodes support parametric workflow construction.
* [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) — expose mesh workflows as callable APIs for Unity and Unreal integration.
* [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices) — backend selection, input preparation, and post-generation pipelines.
* [Animation Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes/animation-nodes) — character rigging and animation workflows that build on Rig Humanoid Mesh and Animate Rigged Model.

## Frequently asked questions

**Which mesh node should I start with?**

For most use cases, start with Image → 3D and follow [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices). For character sheets with multiple views, start with Multi-View → 3D. For production UGC where reliability is critical, use Image → 3D (With Fallback).

**How do I make AI-generated meshes game-ready?**

Standard chain: generate → Auto Transform Mesh → Optimize Mesh (for polycount reduction and topology cleanup) → texture or bake high-to-low → Apply Textures to Mesh. For characters, add Rig Humanoid Mesh + Animate Rigged Model at the end.

**Can I get quad-based meshes for sculpting?**

Yes. Select a quad-based backend in Image → 3D, or use Optimize Mesh with Quad topology setting. Quads are preferred for sculpting workflows in ZBrush, Blender, or Maya; triangles are standard for real-time engines.

**What's the difference between Omnipart and Seperate Object Parts?**

Separate Object Parts takes an existing 3D model (GLB) and slices it into individual components. OmniPart takes a 2D concept image (PNG/JPG), segments it, and generates a brand new 3D model out of those pieces.

* **Use&#x20;*****Separate Object Parts*****&#x20;when:**\
  You already have a 3D model (e.g., a chest, gun, or vehicle) and need to disassemble it into individual parts (lids, wheels, doors) to animate them or make them interactable in a game engine.
* **Use&#x20;*****OmniPart*****&#x20;when:**\
  You only have a 2D concept drawing of an object and want to generate a brand new 3D model that is *already pre-built* in modular, exploded, or segmented parts directly from the 2D image.

**Can mesh nodes export FBX or USD?**

The platform's native output is GLB. For other formats, use external conversion tools (Blender, FBX SDK) downstream of the workflow, or wire the API Output into your engine and convert on import.


# 3D Generation Best Practices

Generate 3D meshes from images in Atlas AI Studio: input prep, backend selection, transform and texture post-processing, and high-poly to low-poly baking.

Each game has unique constraints, but the patterns for generating production-quality 3D assets in Atlas are consistent. This page covers how to prepare inputs, choose the right backend model, route outputs into the rest of your pipeline, and avoid the most common mistakes when generating 3D meshes from a single image.

## When to use the Image to 3D node

Reach for Image to 3D when you have a reference image (concept art, photograph, or 2D render) and need a corresponding 3D mesh as the starting point for a production asset. Common use cases:

* **Concepting and prototyping.** Generate a rough 3D version of a 2D concept in under a minute to evaluate proportions, silhouette, and scale before committing modeling time.
* **Asset library expansion.** Convert existing 2D library art into 3D variants for use in game scenes, marketing material, or virtual production.
* **Hero asset starts.** Produce a high-detail mesh that a technical artist then refines, retopologizes, or bakes down to a game-ready low-poly target.
* **Stylized object generation.** Use stylization-aware backends to produce meshes that respect a specific art direction rather than defaulting to generic realism.

If you need a multi-view-conditioned mesh (front, back, and side reference), pair Image to 3D with the **Multi-view Generation** capability in the upstream image nodes for stronger geometric accuracy on occluded surfaces.

## Image **→** 3D Node

The **Image to 3D** node converts a single image into a fully generated **3D mesh**.\
This node is most effective when the input image is clean, isolated, and visually informative.

### Preparing the Input Image <a href="#id-1-preparing-the-input-image" id="id-1-preparing-the-input-image"></a>

For the best 3D results, the source image should:

* show **only the desired object**
* have a **neutral background** (white, gray, or solid color)
* include **depth cues** such as shadows
* preferably use a **perspective camera angle**
* avoid clutter, overlapping objects, or busy patterns

To achieve this, you can first process the original scene using:

* **Multimodal Node** → isolate the object with neutral background
* **Precise Text Edit Node** → remove all other elements cleanly

<figure><img src="/files/3GkEi2cbMRlewQyO2oOy" alt="" width="563"><figcaption></figcaption></figure>

These steps help the 3D model generator correctly infer:

* thickness
* volume
* proportions
* silhouette
* material hints

### Backends & Output Variation <a href="#id-2-backends--output-variation" id="id-2-backends--output-variation"></a>

The Image to 3D node includes **multiple backend models**, each producing different types of meshes:

#### Backend Differences <a href="#backend-differences" id="backend-differences"></a>

* **High-detail models** → strong surface accuracy
* **Low-detail models** → simplified forms for fast iteration
* **Low-poly models** → game-ready stylized assets
* **Quad-based meshes** → ideal for sculpting or retopology
* **Triangle-based meshes** → standard for real-time engines
* **PBR-enabled meshes** → includes color + material maps
* **Mesh-only output** → no textures, only geometry

<figure><img src="/files/P7BqL8l288eWjyNGj3YF" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/B0EcNSBYBM5B0Z5gN7zm" alt="" width="563"><figcaption></figcaption></figure>

You can toggle:

* **Disable PBR**
* **Disable Texture** for geometry only output.

This helps adapt the output to different pipelines (Blender, Unreal, Unity, CAD, etc.).

<figure><img src="/files/U4z5VO7AteFPBn2qMe8B" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/JSZR6Pd6zADvFqcKmxkF" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/mjS96uRmXkO7phL8uUbD" alt="" width="563"><figcaption></figcaption></figure>

### Generating the Model <a href="#id-3-generating-the-model" id="id-3-generating-the-model"></a>

Once the image and settings are ready:

1. Connect your isolated image to the **Image Input** of the Image to 3D node
2. Select backend model
3. Click **Run**

The node returns:

* a **3D preview**
* a wireframe preview
* a downloadable **.glb** model

You can download it directly or refine it using the next stage of mesh nodes.

### After Generation: Next Steps <a href="#id-4-after-generation-next-steps" id="id-4-after-generation-next-steps"></a>

After generating the mesh, you can optionally pass it through:

#### **Auto Transform Mesh** <a href="#auto-mesh-transform" id="auto-mesh-transform"></a>

* fix origin point
* normalize scale
* reorient axes
* center the model
* prepare it for procedural assembly

#### **Texture Mesh** <a href="#texture-mesh" id="texture-mesh"></a>

* add or enhance textures
* refine materials

#### **Blender / Unreal / Unity via API Nodes** <a href="#blender--unreal--unity" id="blender--unreal--unity"></a>

* import for scene assembly
* integrate into procedural systems
* convert to kits or asset libraries

<figure><img src="/files/ddfpqZmDCkv83onzXmOg" alt="" width="563"><figcaption></figcaption></figure>

## Common pitfalls

1. **Busy or cluttered input images.** Multiple objects, complex backgrounds, or overlapping subjects prevent the Image to 3D node from inferring thickness, volume, proportions, and silhouette, producing distorted geometry. Isolate the subject first with the Multimodal Node for a neutral background, or remove other elements with the Precise Text Edit Node.
2. **Flat, dead-on source views.** An image with no perspective and no shadows gives the generator no depth cue, so scale and volume are guessed. Prefer a perspective camera angle and include shadows.
3. **Choosing a backend that mismatches the target pipeline.** High-detail backends produce dense meshes that are not game-ready without retopology; low-poly backends produce stylized geometry that will not hold up under hero close-ups. Selecting the wrong one forces a re-run to get usable topology.
4. **Toggling Disable PBR or Disable Texture after deciding you need maps.** If the asset requires color and material maps, generate with PBR enabled on the first pass; discovering the need afterward means re-running the generation.
5. **Feeding a raw generated mesh into procedural assembly.** Generated meshes often have non-standard pivot, scale, and orientation, causing misplacement in scenes. Pass them through Auto Transform Mesh first.
6. **Baking onto a low-poly mesh with overlapping UVs.** Bake High-Poly to Low-Poly requires non-overlapping UVs in the 0–1 space; overlapping UVs corrupt the baked normal and PBR channels.

## Related nodes

* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — uploading reference images and art bibles into your workflow
* [Image Generation Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-generation-nodes) — generating concept art images that feed the Image to 3D node
* [Image Edit Nodes](/atlas-ai-studio-overview/node-index/image-nodes/image-edit-nodes) — refining and isolating subjects before 3D generation
* [2D Post Processing Nodes](/atlas-ai-studio-overview/node-index/image-nodes/2d-post-processing-nodes) — preparing clean inputs through background removal, masking, and recoloring
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — overview of all mesh-related capabilities, including Texture Mesh and Bake High-Poly to Low-Poly
* [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) — exporting a finished workflow as a callable API for scene assembly in Blender, Unreal, or Unity

## Frequently asked questions

**What input image resolution works best?**

A square or near-square image at 1024×1024 pixels or higher is ideal. Lower resolutions degrade detail recovery in the generated mesh; higher resolutions do not usually improve output and slow down inference. The source image should show only the desired object on a neutral background such as white, gray, or a solid color.

**Can the Image to 3D node handle transparent backgrounds?**

Yes. Transparent PNG inputs work as expected. The generator treats transparent regions as empty space, which produces cleaner silhouettes than colored backgrounds. If your input has a complex background, isolate the subject first with the Multimodal Node.

**Which backend is best for game-ready output?**

For real-time engine targets such as Unity and Unreal, start with the low-poly or stylized backends, which produce game-ready geometry directly. For hero assets where polycount matters less, high-detail backends give better surface accuracy. To combine both, generate a high-detail mesh, retopologize to a low-poly target, then use the Bake High-Poly to Low-Poly node to transfer detail by baking normal, albedo, roughness, metallic, and ambient occlusion maps onto the low-poly mesh. The low-poly mesh must have non-overlapping UVs in the 0–1 space, and you set an explicit texture resolution for the baked maps.

**Why does the back of my generated mesh look incorrect?**

Single-image input gives the generator only one view, so the back, top, and occluded surfaces are inferred rather than observed. Provide a cleaner, well-lit source image with depth cues to improve inference of thickness and volume on unseen surfaces.

**Can I generate quad-based meshes for sculpting?**

Yes. Select a quad-based backend in the Image to 3D node. Quad-based meshes are suited to sculpting and retopology. For real-time rendering, triangle-based meshes are typically more efficient.

**What does the Image to 3D node output?**

The node returns a 3D preview, a wireframe preview, and a downloadable `.glb` model. You can toggle Disable PBR and Disable Texture for geometry-only output when you only need the mesh without color or material maps.

**How do I use the generated mesh in Unreal, Unity, or Blender?**

The output is a standard `.glb` file. Either download it directly and import via the engine's native asset pipeline, or wire your workflow into an [API Node](/atlas-ai-studio-overview/node-index/api-nodes) so Unreal, Unity, or Blender can request the mesh for scene assembly and procedural systems. The API approach keeps the workflow versioned and deterministic for production pipelines.


# Animation Nodes

Animation nodes that rig and animate humanoid character meshes: automatic rigging, Mixamo retargeting, Smart Mesh Alignment, and FBX/GLB export.

Animation Nodes are the bridge between generated 3D character meshes and game-ready animated assets. They handle rigging (applying a skeleton to a static mesh), retargeting (mapping animation data onto the rigged skeleton), and full character pipelines that automate the entire concept-art-to-animated-NPC sequence.

## When to use animation nodes

* **Generate playable NPCs from concept art.** Use the Character Workflow node to chain Split Character Sheet → Multi-View to 3D → Rig Humanoid Mesh → Retarget into one automated pipeline. Single-shot from drawing to animated character.
* **Bring AI-generated humanoid characters into a game engine.** Use Rig Humanoid Mesh followed by Animate Rigged Model to skeleton + animate a generated mesh ready for Unity or Unreal import.
* **Retarget Mixamo animations.** Pair Input Mixamo Animation with the retargeting nodes to apply Mixamo's motion library onto generated characters without manual rigging.
* **Smart-place props on characters.** Use Smart Mesh Alignment with natural-language instructions to attach weapons, accessories, or environmental objects to character rigs without manual transform math.
* **Rig and animate Tripo-generated characters.** Tripo-specific rigging and animation nodes are tuned for Tripo's mesh topology.

### Character Workflow

A composite workflow that converts a multi-view character drawing into a rigged, animated 3D mesh ready for game integration. The pipeline chains Split Character Sheet, Multi-View to 3D, Rig Humanoid Mesh, Retarget Rigged Mesh, and Retarget Mixamo Animation into a single automated process.

<figure><img src="/files/dFbO9LuVsmSKaPsfWy1S" alt="" width="563"><figcaption></figcaption></figure>

* **Input**: A character sheet image showing front, side, and back views of a character design.
* **Split Character Sheet**: Automatically extracts individual orthographic views from the input sheet.
* **Multi-View to 3D**: Generates a 3D mesh by interpreting the separated views.
* **Rig Humanoid Mesh**: Applies a skeleton armature suitable for humanoid characters.

<figure><img src="/files/ZkfDHmEBQJOgI7Or6HmW" alt="" width="357"><figcaption></figcaption></figure>

<figure><img src="/files/pv7w8Auqup9ZTPqWGicQ" alt="" width="563"><figcaption></figcaption></figure>

* **Retarget Rigged Mesh**: Adjusts the rig to a T-pose or reference pose.
* **Retarget Mixamo Animation**: Maps animation clips (e.g., from Mixamo) onto the generated rig.
* **Output**: A fully rigged, animated character mesh exported in standard 3D formats.

Useful for rapid prototyping of NPC models, testing character designs in-engine, and generating background characters from concept art.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fnt3ZgWr7XkufTfAB0WVx%2Frig%20animate.mp4?alt=media&token=c1605604-2bc3-4c99-a455-8bf33930592b>" %}

Smart Mesh Alignment

Automatically positions and rotates one 3D mesh onto another using natural language instructions. Functions as an AI-powered spatial alignment tool that interprets placement intent without manual transform adjustments.

<figure><img src="/files/Hzow3NGlYwAdalXPSNmb" alt="" width="563"><figcaption></figcaption></figure>

* **Mesh to Align**: The 3D object to be repositioned—props, accessories, attachments, or environmental details.
* **Anchor Mesh**: The reference geometry that defines the target location and orientation.
* **Text Instruction**: Natural language description of the desired placement (e.g., "place on top of the table," "attach to the character's right hand").
* **Automatic Transform**: Calculates position and rotation based on the instruction and anchor geometry, interpreting spatial relationships like "above," "inside," or "next to."
* **Output**: The aligned mesh with updated transform data, ready for scene composition or further editing.

Useful for rapidly attaching weapons to character rigs, positioning props in environment layouts, snapping accessories to vehicle models, and assembling modular level pieces from text descriptions.

### Animate Nodes

A pair of connected nodes that rig and animate AI-generated humanoid meshes. The Rig Humanoid Mesh node applies a skeletal armature to a 3D character model, while the Animate Rigged Mesh node applies preset animation clips to the rigged result.

<figure><img src="/files/mkz7Gl35EDBKTWUcgdp1" alt="" width="563"><figcaption></figcaption></figure>

* **Rig Humanoid Mesh Input**: Accepts a 3D human mesh, with optional texture image and character height parameters for scale calibration.
* **Skeleton Generation**: Automatically places a humanoid bone hierarchy (spine, limbs, head) based on mesh topology.
* **Animate Rigged Mesh**: Takes the rigged output and applies selectable animation presets (walk, run, idle, combat, etc.).
* **Animation Selection**: Provides a dropdown or picker interface for choosing from built-in motion clips.
* **Output**: A rigged and animated 3D character ready for export in standard formats (FBX, GLB).
* **Pipeline Position**: Typically follows mesh generation nodes and precedes export or retargeting stages.

Useful for quickly animating procedurally generated NPCs, testing character proportions with motion, and prototyping dialogue or cutscene performances.

### Tripo Mesh Nodes

A paired set of nodes designed to rig and animate 3D character meshes generated by Tripo workflows. The Rig Tripo Mesh node applies a skeletal armature to the model, while the Animate Tripo Mesh node attaches preset animation clips to the rigged character.

<figure><img src="/files/vSBBQihPjknyJ48KVF6r" alt="" width="445"><figcaption></figcaption></figure>

<figure><img src="/files/hr192e1KwYjW9gNz3Pxt" alt="" width="563"><figcaption></figcaption></figure>

* **Rig Tripo Mesh**: Accepts a Tripo-generated mesh and applies a skeleton with bone hierarchy and automatic weight painting.&#x20;
* **Multi-Creature Support**: Auto-detects or configures skeletons for **Bipeds** (humans), **Quadrupeds** (four-legged animals), **Hexapods/Octopods** (insects/spiders), **Serpentine** (snakes), **Avians** (birds), and **Aquatics** (fish).
* **Animate Tripo Mesh**: Takes the rigged output and applies preset animation clips such as idle, walk, or combat cycles.
* **Sequential Workflow**: The rig node must precede the animation node in the pipeline; animations require an existing skeleton structure.
* **Preview Support**: Both nodes provide real-time viewport preview of the rigged mesh and active animation playback.
* **Export Ready**: Output includes both geometry and animation data in standard 3D formats for engine integration.

Useful for rapidly prototyping animated NPCs, testing character movement systems, and generating background characters from Tripo-created models.

### Tripo Mesh Workflow

A composite workflow that processes Tripo-generated 3D meshes through rigging and animation stages to produce game-ready animated characters. The pipeline integrates mesh preparation, skeletal rigging, and animation retargeting into a single automated sequence.

<figure><img src="/files/ZrzX7wBeup6Uw5wFfCdp" alt="" width="563"><figcaption></figcaption></figure>

* **Input**: A 3D mesh of a character preferably in T-pose or A-pose, typically from text or image prompts.
* **Rig Tripo Mesh**: Applies a skeletal armature to the input mesh, automatically detecting joint positions and assigning bone weights.
* **Animate Tripo Mesh**: Retargets animation clips onto the rigged character, adjusting for the specific bone structure and proportions.
* **Animation Library**: Supports standard animation types including locomotion cycles, idle poses, and action sequences.
* **Bone Mapping**: Automatically maps generic animation data to the Tripo mesh's unique skeleton hierarchy.
* **Output**: An animated 3D character with embedded skeleton and animation data, ready for export.

Useful for rapidly prototyping animated NPCs from text descriptions, testing character concepts with motion, and generating background characters with basic animation cycles.

## Common pitfalls

* **Skipping the height parameter on Rig Humanoid Mesh.** Without an accurate character height value, joint positions are misscaled and the applied animation looks off.
* **Rigging a mesh in an action or dynamic pose.** Skeleton placement is derived from mesh topology, so non-standard poses cause joint placement errors. Regenerate the source mesh in T-pose or A-pose before rigging.
* **Running Animate Tripo Mesh before Rig Tripo Mesh.** Animation nodes attach clips to an existing skeleton, so the rig node must run first; reversing the order leaves nothing for the animation to bind to.

## Related nodes

* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — upstream generation, transformation, and optimization nodes that produce the meshes fed into animation workflows.
* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — Input Mixamo Animation supplies external animation clips for Retarget Mixamo Animation to map onto generated characters.
* [Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes) — generate or refine the character art that feeds Split Character Sheet and Multi-View to 3D inside the Character Workflow.
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — Split Character Sheet prepares multi-view character sheets for 3D generation; LLM nodes can parametrize animation selection from text.
* [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices) — backend and input-prep guidance affecting the starting mesh an animation workflow receives.

## Frequently asked questions

**Can I animate non-humanoid characters?**

Yes. While Rig Humanoid Mesh expects humanoid topology, Rig Tripo Mesh can be used for different creatures like **Quadrupeds** (four-legged animals), **Hexapods/Octopods** (insects/spiders), **Serpentine** (snakes), **Avians** (birds), and **Aquatics** (fish).

**What animation clips can I apply?**

Animate Rigged Mesh applies selectable preset clips from a built-in library covering locomotion (walk, run), idle poses, and combat cycles. For Mixamo clips, supply an FBX through Input Mixamo Animation and let Retarget Mixamo Animation map it onto the rigged character.

**Can I use custom keyframe animation?**

The animation nodes are preset-driven. Author custom keyframe animation in external tools and bring it in through Input Mixamo Animation for retargeting.

**What format does the rigged, animated character export in?**

FBX and GLB. The export includes vertex geometry, UV mapping, material assignments, the humanoid skeleton with joint hierarchy and bone weights, and the retargeted animation data.

**Why does my character look distorted after rigging?**

The two usual causes are a source mesh that is not in a standard pose and an incorrect height value. Regenerate the mesh in T-pose or A-pose and set the Rig Humanoid Mesh character height to the intended in-game scale.

**Can Smart Mesh Alignment attach props to a specific bone?**

No. Smart Mesh Alignment positions and rotates one mesh onto an anchor mesh from a natural-language instruction and anchor geometry, interpreting relationships like "above," "inside," or "next to." It suits one-off placement during workflow design. For bone-specific socket attachment at runtime, use the engine's native socket system after import.

**How do I rig and animate a Tripo-generated mesh?**

Use the Tripo Mesh Nodes pair or the Tripo Mesh Workflow. Rig Tripo Mesh applies a humanoid skeleton with automatic weight painting, then Animate Tripo Mesh applies preset clips. The Tripo Mesh Workflow combines both stages into one automated sequence and maps generic animation data to the mesh's skeleton hierarchy.

**What does the Character Workflow produce from a character sheet?**

From a sheet showing front, side, and back views it produces a rigged, animated 3D mesh, chaining Split Character Sheet to extract the views, Multi-View to 3D for geometry, Rig Humanoid Mesh for the skeleton, Retarget Rigged Mesh to set a reference pose, and Retarget Mixamo Animation to map clips.

**What is the difference between Rig Humanoid Mesh and Rig Tripo Mesh?**

Rig Humanoid Mesh (Meshy Engine): Strictly for humans/bipeds (T-pose/A-pose) and includes built-in walking/running preview animations. Rig Tripo Mesh (Tripo Engine): For any creature shape (quadrupeds, birds, spiders, fish, snakes) as well as humans.

* **Use&#x20;*****Rig Humanoid Mesh*****&#x20;for**: Standard human characters, heroes, enemies, or humanoids.
* **Use&#x20;*****Rig Tripo Mesh*****&#x20;for**: Animals, monsters, monsters with wings, spiders, or any non-humanoid creatures.


# Video Nodes

Video Nodes generate video from text and reference images, and edit, extend, lipsync, and upscale existing clips.

Video Nodes generate short animated clips directly from text prompts and optional reference images.\
They are ideal for creating **cinematic previews**, **marketing videos**, **quick scene animations**, or **visual storytelling elements** within Atlas workflows.

Two nodes are available: a full-featured version and a simplified fast-generation version.

## When to use video nodes

Reach for video nodes when a static image isn't enough and a fully-rigged animation pipeline is overkill. Common use cases:

* **Marketing and trailers.** Generate cinematic clips from concept art for pitch decks, store listings, social media, or community announcements.
* **Cutscene prototyping.** Roughly visualize a cinematic before committing animator time to a polished version.
* **Animated moodboards.** Turn a single style reference into a short looping clip that conveys mood and motion direction to the team.
* **NPC and character animation prototyping.** Use Lipsync to put a generated voice on a static character portrait for dialogue review, or Reference to Video to test motion choreography against a character reference.
* **Video edits and continuations.** Use Video Edit to restyle existing footage and Video Extend to grow short clips into longer sequences without re-rendering.

### Text + Image -> Video <a href="#text--image-to-video" id="text--image-to-video"></a>

Generates a video from a **text prompt**, an optional **continuation prompt**, and up to **three reference images**.

<figure><img src="/files/GB8qJAEHfJi9t9vtSUUc" alt="" width="563"><figcaption></figcaption></figure>

#### Inputs <a href="#inputs" id="inputs"></a>

* **Prompt** — main instruction for video content
* **Continuation Prompt** (optional) — describes how the animation should evolve
* **Reference Images** — up to **3** images to control style, subject, or composition

#### Video Settings <a href="#video-settings" id="video-settings"></a>

* **Duration:** 4, 6, 8, 16, 24, 32, or 40 seconds
* **Resolution:** 720p or 1080p
* **Aspect Ratio:** Landscape or Portrait

#### Output <a href="#output" id="output"></a>

* A rendered video clip in the chosen format

This node is suited for more controlled, style-specific video generation, especially when reference images are important.

#### Example Usecase <a href="#example-usecase" id="example-usecase"></a>

### Simple Text + Image -> Video <a href="#text--image-to-video-simple" id="text--image-to-video-simple"></a>

A streamlined version optimized for **fast, lightweight video generation**.

<figure><img src="/files/WQkvS0VC8zFlMRa2W1uG" alt="" width="563"><figcaption></figcaption></figure>

#### Inputs <a href="#inputs-1" id="inputs-1"></a>

* **Prompt** — primary description
* **Input Image** (optional) — style or subject reference

#### Video Settings <a href="#video-settings-1" id="video-settings-1"></a>

* **Duration:** 5, 10, or 12 seconds
* **Resolution:** 420p, 720p, or 1080p
* **Aspect Ratio:** Landscape, Portrait, Standard, or Square
* **Fixed Camera Position:** enable or disable
* **Seed:** control variation (`-1` = random)

#### Output <a href="#output-1" id="output-1"></a>

* A quick-rendered video clip

This version is ideal for rapid prototyping or generating simple animated assets for marketing or social media.

<figure><img src="/files/q3vBp5lPr6iIqwGBRaDE" alt="" width="540"><figcaption></figcaption></figure>

#### Example Usecase <a href="#example-usecase-1" id="example-usecase-1"></a>

* **End Image** input — provides control over both the start and end frames of the generated video clip.

<figure><img src="/files/zJVkDPIo0lp9NILB0O0V" alt="" width="563"><figcaption></figcaption></figure>

### Use Cases <a href="#use-cases" id="use-cases"></a>

* Marketing videos from a single concept image
* Animated moodboards
* Scene previews for game or environment design
* Quick animations for pitch decks or client presentations
* Stylized loops for social media

Video Nodes provide a fast way to bring static concepts to life using text prompts and reference imagery.

### Video Edit

Transforms an existing video by applying a new creative direction or environment using a reference image and text prompt.

<figure><img src="/files/LovEP1B8IZZzRazmKovt" alt="" width="563"><figcaption></figcaption></figure>

**Inputs**

* **Source Video** — the original video clip to transform
* **Reference Image** — visual guide for the target style, environment, or look
* **Prompt** — text description of the desired edit

**Parameters**

* **Backend selector** — choose motion-path generation method (some backends use prompt-based motion, others use reference-driven paths)
* **Seed** — control variation (`-1` = random)

**Output**

* Edited video clip matching the reference image style and prompt direction

**Useful for:** changing the setting or atmosphere of placeholder footage, creating environmental variations of cutscenes, or adapting generic vehicle or character animations into themed game contexts (expedition tours, combat zones, fantasy landscapes).

* Accepts input videos from 3 to 60 seconds in length
* Supports up to 5 reference images
* Output resolution selectable as 720p or 1080p
* Audio handling mode: automatic or original (preserve source audio)
* Some backends support instruction-based edits with optional style reference image
* Supports style transfer driven by a reference video in addition to reference images

<figure><img src="/files/LovEP1B8IZZzRazmKovt" alt="" width="563"><figcaption></figcaption></figure>

* Supports placing synchronized sound effects onto a video

### Video Extend

Extends an existing video clip forward in time by generating additional frames based on a text prompt and the final frames of the input.

<figure><img src="/files/zTA2J36h0Yzo0Xe5KN2G" alt="" width="563"><figcaption></figcaption></figure>

**Inputs**

* **Source Video** — the video clip to continue
* **Prompt** — text description guiding the extended footage
* **Backend selector** — choose generation engine (different backends produce varying motion styles and continuation approaches)

**Parameters**

* **Duration** — length of the extension (available durations depend on the selected backend)
* **Resolution** — output resolution (options vary by backend)
* **Seed** — control variation (`-1` = random)

**Output**

* Extended video clip appended to the original

**Useful for:** creating longer cutscene sequences from short generated clips, looping environmental footage, or prototyping extended NPC actions and vehicle animations without re-rendering the entire scene.

### Seedance Reference to Video

Generates video content by combining multiple reference inputs—images, video clips, and audio—with a text prompt to produce a cohesive animated result.

<figure><img src="/files/Ru5aNel1obDsDPSG4SQk" alt="" width="563"><figcaption></figcaption></figure>

**Inputs**

* **Prompt** — text description guiding the generation
* **Reference Images** — up to 9 still images for style, character, or environment guidance
* **Reference Videos** — up to 3 video clips (e.g., motion choreography, background loops, camera movement)
* **Reference Audio** — up to 3 audio clips to influence pacing, rhythm, or mood

**Parameters**

* **Seed** — control variation (`-1` = random)

**Output**

* Generated video clip synthesizing all provided references

<figure><img src="/files/6yYA0VzmO3ZM4YSk15GT" alt="" width="563"><figcaption></figcaption></figure>

**Useful for:** creating NPC dance sequences synced to in-game music, generating character performances driven by reference choreography, or producing cutscene animations that blend concept art, motion samples, and soundtrack cues.

### Lipsync

Synchronizes a character's mouth movements to match an audio track, producing a video of the character speaking the provided dialogue or narration.

<figure><img src="/files/TQ8n8turlAGYcjV9tfOO" alt="" width="563"><figcaption></figcaption></figure>

* **Video Input** — the character or face video to animate
* **Audio Input** — the speech or dialogue track to sync
* **Backend Selection** — four selectable backends, each with different processing characteristics
* **Output** — a lip-synced video clip with matching mouth movements
* Requires a clearly visible speaker facing the camera during the audio timeframe
* Best used with close-up or medium shots where facial features are distinct

<figure><img src="/files/8sEmQzURf9JKxkMt8VHB" alt="" width="563"><figcaption></figcaption></figure>

### Upscale Video

Increases the resolution of an input video to produce sharper, higher-fidelity output from lower-resolution source clips.

<figure><img src="/files/reCvq0mQIIULwZKU37TK" alt="" width="563"><figcaption></figcaption></figure>

* **Video input.** Accepts an existing clip as the source for upscaling.
* **Video output.** Returns the upscaled clip at increased resolution.
* **Resolution ceiling.** Upscales output up to 1080p at 30fps.
* **Preserves timing.** Frame count and duration of the source are retained; only spatial resolution is increased.
* **Typical use.** Finalize prototyped or generated footage for trailers, store listings, and cutscene review at higher fidelity.

Useful for: promoting rough or low-resolution generated clips to presentation quality without re-rendering the original video.

## Common pitfalls

* **Mismatched aspect ratio across reference images.** When supplying multiple reference images to the Text + Image -> Video node, references with very different aspect ratios produce inconsistent framing. Pre-crop references to a consistent shape via 2D Post Processing before generation.
* **Expecting backends to share durations and resolutions.** Different backends in the Video Extend node support different duration and resolution options, and backend-driven motion in the Video Edit node varies similarly. Check the dropdown per node rather than assuming parameter parity.
* **Feeding Lipsync unclear audio.** The Lipsync node produces visibly worse mouth movement from background noise, music behind dialogue, or low-quality recordings. Clean or generate the speech track via Audio Nodes first.
* **Feeding Video Edit a source outside its length range.** The Video Edit node accepts input videos from 3 to 60 seconds; clips outside that range cannot be transformed.
* **Expecting Upscale Video to add motion or length.** The Upscale Video node increases spatial resolution only and retains the source frame count and duration; it does not regenerate content or extend length. Use the Video Extend node for additional duration.

## Related nodes

* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — Input Image and Input Images supply reference content for video generation; Input Video supplies source clips for the Video Edit and Video Extend nodes.
* [Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes) — generate or refine reference frames before producing video.
* [Audio Nodes](/atlas-ai-studio-overview/node-index/audio-nodes) — generate voice tracks for the Lipsync node and audio clips used as reference audio in the Seedance Reference to Video node.
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — combine multiple references, prompts, or extracted document content into video-ready inputs.
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — provide rigging and animation workflows for real-time character animation, an alternative to pre-rendered video clips.

## Frequently asked questions

**What is the difference between the Text + Image -> Video and Simple Text + Image -> Video nodes?**

The Text + Image -> Video node accepts more parameters—a continuation prompt, up to three reference images, and longer durations—for more controlled, style-specific output. The Simple Text + Image -> Video node is a streamlined text to video AI node with one optional input image and fewer settings, optimized for fast, lightweight generation. Use the Simple node for rapid prototyping and the full node when reference images and control matter.

**How long can generated videos be?**

Duration depends on the node. The Text + Image -> Video node supports 4, 6, 8, 16, 24, 32, or 40 seconds. The Simple Text + Image -> Video node supports 5, 10, or 12 seconds. The Video Extend node appends additional duration to an existing clip, with available durations depending on the selected backend. For longer sequences, chain Video Extend calls.

**Can I generate video with sound?**

The video generation nodes can produce visual output with baked-in audio according to the backend. For audio, you can also generate it separately via [Audio Nodes](/atlas-ai-studio-overview/node-index/audio-nodes) and combine it in post-production, or use the Lipsync node to synchronize generated dialogue onto a character video directly.

**Why does my video look stylistically different from my reference image?**

Some video generation backends prioritize motion fidelity over strict style adherence. If style consistency matters, use the Seedance Reference to Video node, which accepts up to nine reference images, three reference videos, and three reference audio clips, or the Text + Image -> Video node rather than the Simple Text + Image -> Video node.

**Can I use video output as a real-time engine asset?**

Generally no. Video outputs are pre-rendered clips, not interactive content. Use them for cinematics, marketing material, in-engine playback such as UI screens and in-world displays, or as reference for traditional animation. For real-time character animation, use the rigging workflows in [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) instead.

**What is the recommended workflow for animated NPC dialogue?**

For lipsync generation, generate the speech track via [Audio Nodes](/atlas-ai-studio-overview/node-index/audio-nodes) text-to-speech, then feed it into the Lipsync node along with a character video. The Lipsync node offers four selectable backends and synchronizes mouth movement to the audio; it requires a clearly visible speaker facing the camera and works best with close-up or medium shots. For broader body animation, use the rigging and animation pipeline in [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes).

**How do I set the start and end frames of a generated clip?**

The Simple Text + Image -> Video node accepts an optional Input Image and an End Image input, providing control over both the start and end frames of the generated video clip.

**How do I change the setting or style of existing footage?**

Use the Video Edit node, which transforms a source video using a reference image and a text prompt. It suits changing the atmosphere of placeholder footage, creating environmental variations of cutscenes, or adapting generic animations into themed game contexts. A backend selector chooses the motion-path generation method.

**How do I improve the resolution of a generated clip?**

Use the Upscale Video node. It increases the resolution of a lower-resolution source up to 1080p at 30fps while retaining the original frame count and duration, which finalizes prototyped or generated footage for trailers, store listings, and cutscene review without re-rendering.

**What is the difference between Seedance 2.0 Reference to Video and Text + Image → Video?**

Seedance 2.0 Reference to Video: A highly multimodal node. It lets you link multiple images (up to 9), videos (up to 3), and audio clips (up to 3) together in your prompt (e.g., matching a face to a voice reference). Text + Image → Video: A multi-engine playground. Best for high-fidelity animations, smooth morphing from a start image to an end image, or writing complex multi-shot timelines.

* **Use&#x20;*****Seedance 2.0 Reference to Video*****&#x20;when**: You need to synthesize video, motion, and audio (like matching a character portrait to a specific voiceover/sound reference).
* **Use&#x20;*****Text + Image → Video*****&#x20;when**: You want maximum cinematic quality, need to animate a single image, or want to morph a start image into a different end image.


# Audio Nodes

Atlas AI Studio audio nodes generate speech, music, and sound effects from text, with multi-speaker dialogue, voice changing, and transcription.

Audio Nodes are how Atlas generates and processes audio content: dialogue, music, sound effects, and voice transformations. Most game audio pipelines start with reference recordings, voice actors, or licensed libraries. Atlas audio nodes let teams generate placeholder and production-tier audio directly from text prompts, then transform and route it through standard game audio formats (MP3, WAV).

## When to use audio nodes

* **Voiceover prototyping.** Generate placeholder dialogue for NPCs, cutscenes, tutorials, and quest briefings during pre-production, before committing to professional voice actors.
* **Multi-character conversation generation.** Use Text to Dialogue or Audio Speech Workflow to produce full multi-speaker conversations with distinct voices per character.
* **Dynamic music systems.** Generate background music, combat themes, ambient soundscapes, and menu music from text descriptions. Useful for procedural audio systems where pre-rendered tracks don't fit every scenario.
* **Sound effects from text.** Generate one-shot effects, UI sounds, and environmental audio without sound library licensing.
* **Voice variety from a single recording.** Apply Voice Changer to a single source voice and produce multiple character variants for NPC barks, system announcements, or background chatter.

### Audio Speech Workflow

A pre-built workflow template for generating character dialogue audio with voice assignment and optional voice transformation. Combines text-to-speech synthesis with voice tagging and modification to produce character-specific speech output in MP3 format.

<figure><img src="/files/tM0lUdIema2SRJwiGPQo" alt="" width="563"><figcaption></figcaption></figure>

* **Add Audio Tags** node parses dialogue lines in `A:` / `B:` format and assigns distinct voices to each speaker
* **Text-to-speech** nodes convert tagged dialogue into synthesized audio using selectable backend models
* **Voice Changer** node applies transformations (pitch, timbre, effects) while maintaining speech timing and emotional tone
* Output format is MP3, suitable for direct integration into game audio pipelines
* Dialogue rhythm and pacing are preserved through voice modification stages
* Supports multiple speakers in a single workflow for conversation sequences

**Useful for:** NPC dialogue generation, cutscene voiceover prototyping, dynamic conversation systems, placeholder voice acting during development.

### Audio Nodes Overview

A collection of nodes for generating, transforming, and processing audio content within AI workflows. Enables synthesis of speech, music, and sound effects from text or other inputs, with support for voice manipulation, format conversion, and multi-speaker dialogue systems.

<figure><img src="/files/O7YBXp3uGtBT7elMZKlG" alt="" width="563"><figcaption></figcaption></figure>

* **Text-to-speech nodes** convert written dialogue and narration into spoken audio using selectable backend models
* **Voice modification nodes** apply pitch, timbre, and effect transformations to generated or imported audio
* **Audio tagging nodes** parse speaker-labeled text formats and route distinct voices to different characters
* **Format conversion nodes** output to common game audio formats (MP3, WAV) for engine integration
* Workflow templates combine multiple audio nodes for common scenarios like dialogue generation and voice acting
* Supports both single-voice synthesis and multi-speaker conversation sequences
* Audio timing and emotional delivery are preserved through processing chains

**Useful for:** Voiceover prototyping, placeholder dialogue during development, NPC barks and ambient speech, dynamic audio content generation, cutscene audio mockups.

### Text to Speech (ElevenLabs)

Converts written text into synthesized speech audio using ElevenLabs voice models. Generates spoken dialogue, narration, or voiceover content from text input with selectable voice characteristics and output format options.

* Accepts plain text input for conversion to speech audio
* Voice selection determines speaker identity, tone, and delivery style
* Output format options include MP3 and WAV for game engine compatibility
* Supports emotional delivery and natural speech patterns based on text content
* Model selection affects voice quality, language support, and synthesis characteristics
* Preserves punctuation-driven pacing and emphasis in spoken output

**Useful for:** Character dialogue prototyping, NPC voice generation, cutscene narration, placeholder voiceover during development, dynamic speech content for procedural dialogue systems.

### Simple Text to Speech

Converts text input into synthesized speech audio with selectable backend models. Provides a streamlined interface for generating voiceover and dialogue without multi-speaker tagging or voice transformation stages.

* **Text input** accepts plain dialogue, narration, or script lines without speaker labels
* **Backend selection** offers multiple text-to-speech models with varying voice quality and synthesis characteristics
* Outputs audio in standard formats compatible with game engine audio systems
* Single-voice synthesis — does not parse or route multiple speakers (use Audio Speech Workflow for multi-character dialogue)
* Suitable for rapid prototyping of voiceover content or generating placeholder audio
* Does not include built-in voice modification — chain with Voice Changer node for pitch or timbre adjustments

**Useful for:** Quick voiceover prototyping, NPC barks and system announcements, tutorial narration, placeholder dialogue during pre-production, ambient voice content.

### Add Audio Tags (ElevenLabs)

Parses dialogue text in speaker-labeled format and inserts voice assignment tags for downstream text-to-speech processing. Recognizes `A:` / `B:` / `C:` prefixes and maps each speaker to a distinct voice identifier, enabling multi-character conversations within a single audio workflow.

<figure><img src="/files/XcQd64Jk6qu3VQqJ5VAU" alt="" width="563"><figcaption></figcaption></figure>

* Accepts plain text input with speaker labels (`A: Hello`, `B: Hi there`) and outputs tagged text ready for synthesis
* Each unique speaker prefix is automatically assigned a different voice identifier
* Supports unlimited speakers within a single dialogue block; speaker assignments persist throughout the text
* Tagged output connects directly to text-to-speech nodes that recognize voice metadata
* Preserves line breaks and dialogue structure while adding voice routing information
* No manual voice configuration required; speaker-to-voice mapping is handled automatically based on label order

**Useful for:** Multi-character cutscene dialogue, NPC conversation sequences, branching dialogue prototyping, automated voice casting for scripted exchanges.

### Text to Dialogue (ElevenLabs)

Generates multi-speaker conversation audio from text, with automatic voice assignment and character-based voice selection. Parses dialogue scripts and synthesizes speech for each speaker using distinct voices, producing conversation-ready audio output.

<figure><img src="/files/1vW5es5hq96pBNi2zaho" alt="" width="563"><figcaption></figcaption></figure>

* **Voice assignment** maps each speaker in the script to a selectable voice from the backend's voice library
* **Character category browser** organizes available voices by archetype (hero, villain, merchant, etc.) for quick casting
* Accepts dialogue text with speaker labels and outputs synchronized multi-speaker audio
* Maintains conversation timing and turn-taking between speakers automatically
* Outputs audio in formats suitable for direct integration into game dialogue systems
* Supports rapid iteration on character voice selection without re-entering dialogue text

**Useful for:** NPC conversation prototyping, cutscene dialogue generation, placeholder voiceover during development, character voice casting exploration, dynamic dialogue system testing.

### Voice Changer (ElevenLabs)

Applies real-time voice transformations to audio input, modifying pitch, timbre, and tonal characteristics while preserving speech timing and intelligibility. Processes synthesized or recorded audio to create character-specific vocal variations.

<figure><img src="/files/d5sltsftSQd2S8kdjv2n" alt="" width="563"><figcaption></figcaption></figure>

* Accepts audio input from text-to-speech nodes or external audio sources
* Transforms vocal characteristics including pitch shift, formant adjustment, and timbre modification
* Background noise removal option cleans audio artifacts during transformation
* Maintains original speech rhythm, pacing, and emotional delivery through processing
* Outputs modified audio in standard formats compatible with game audio pipelines
* Works downstream of voice tagging and text-to-speech nodes in multi-stage workflows
* Speech timing alignment ensures lip-sync compatibility is preserved

**Useful for:** Creating vocal variety for NPCs from a single voice source, transforming placeholder dialogue, generating distinct character voices in conversation sequences, prototype voiceover with limited recording assets.

### Voice Selection

A utility node for browsing, auditioning, and selecting text-to-speech voices from available backend voice libraries. Provides categorized voice lists with metadata (language, gender, style tags) and in-editor audio preview to streamline voice assignment for character dialogue and narration.

<figure><img src="/files/IO5SUObGha6Zi3foATB1" alt="" width="563"><figcaption></figcaption></figure>

* Organizes available voices by category (Narration, Character, Emotional, etc.) and language for quick filtering
* Displays voice metadata including gender, accent, age range, and descriptive style tags
* Built-in audio preview plays sample recordings of each voice directly in the editor without generating new audio
* Selected voice identifier outputs as a string parameter for connection to downstream text-to-speech nodes
* Supports multilingual voice libraries; language filters adapt to the backend's available voice catalog
* Voice availability and categorization depend on the connected backend model

**Useful for:** Character voice casting, dialogue prototyping, matching voice tone to narrative context, rapid iteration on NPC speech styles, placeholder voiceover selection.

### Composition Plan (ElevenLabs)

A node for structuring and organizing multi-segment audio compositions with planned speaker assignments, timing, and content direction. Generates a structured plan that can be consumed by downstream audio generation nodes to produce coherent multi-part audio sequences.

* Accepts text descriptions of audio segments, speaker roles, and sequence structure as input
* Outputs a structured composition plan defining segment order, speaker assignments, and content guidelines
* Enables pre-planning of complex audio sequences before synthesis, separating creative direction from generation
* Plan format is compatible with ElevenLabs audio generation workflows for execution
* Supports multi-speaker scenarios where different voices are assigned to specific segments or roles
* Allows iteration on composition structure without regenerating audio until the plan is finalized

**Useful for:** Planning multi-character dialogue sequences, structuring narrative voiceover with multiple segments, organizing cutscene audio with speaker transitions, prototyping complex conversation flows before full synthesis.

### Create Music (ElevenLabs)

Generates music tracks from text prompts or structured composition plans. Accepts natural language descriptions of musical style, mood, instrumentation, and structure, producing audio output suitable for in-game music, prototyping, and placeholder soundtracks.

<figure><img src="/files/PX2c50nOcTlZGtL0ebLM" alt="" width="563"><figcaption></figcaption></figure>

* Accepts text prompts describing genre, tempo, instrumentation, and emotional tone
* Supports structured **Composition Plan** JSON input for more precise control over musical arrangement and timing
* **Strict duration** toggle enforces exact output length when enabled; relaxed mode allows natural musical phrase endings
* Output varies in length based on strict duration setting (e.g., 2:30 vs 4:00 for the same composition plan)
* Produces audio files ready for integration into game engines
* Works standalone with text prompts or in workflows consuming composition plan data from upstream nodes

**Useful for:** Background music generation, combat themes, ambient soundscapes, menu music, rapid prototyping of audio mood boards, placeholder soundtrack creation during development.

### Simple Music Generation

Generates original music compositions from text descriptions using selectable backend models. Produces audio output suitable for background music, ambient soundscapes, and musical prototyping in game environments.

<figure><img src="/files/2XA6onxqflwoY1XYZx8s" alt="" width="563"><figcaption></figcaption></figure>

* **Text prompt input** describes musical style, mood, instrumentation, and tempo for the desired composition
* **Backend selection dropdown** offers multiple music generation engines with varying style capabilities and output characteristics
* Outputs audio in standard formats compatible with game engine import pipelines
* Generation parameters vary by selected backend; some models support extended duration or specific genre specialization
* Produces royalty-free music assets that can be integrated directly into game builds
* Single-node workflow for rapid iteration on musical concepts without external audio software

**Useful for:** Placeholder background music during development, ambient soundscapes for exploration areas, dynamic music prototyping, mood-based audio testing, quick musical mockups for cutscenes.

### Simple Modify Music

Transforms existing music audio through four modification modes: generating vocal lyrics over instrumental tracks, inpainting specific time segments, remixing with style changes, or creating cover versions with different instrumentation or vocals.

<figure><img src="/files/oIv3roBkdMHzuquJyehK" alt="" width="563"><figcaption></figcaption></figure>

* **Lyrics Only mode** adds sung vocals to instrumental music using provided text prompts
* **Inpainting mode** regenerates a specified time range within the track while preserving surrounding audio continuity
* **Remix mode** restructures the existing composition with tempo, arrangement, or stylistic variations
* **Cover mode** recreates the track with alternative genre, instrumentation, or vocal interpretation
* Accepts audio input in common formats alongside text prompts describing the desired modification
* Preserves musical structure and timing where appropriate to the selected mode
* Output suitable for adaptive music systems, combat variations, and dynamic soundtrack prototyping

**Useful for:** Creating alternate versions of combat music, generating vocal variants of existing themes, prototyping adaptive soundtrack transitions, producing genre variations of menu music.

### Sound Effects (ElevenLabs)

Generates sound effects from text descriptions using ElevenLabs audio synthesis. Takes a written prompt describing the desired sound and outputs an audio file suitable for game integration.

<figure><img src="/files/3tbMUVcs4XDpyV2PwmGS" alt="" width="563"><figcaption></figcaption></figure>

* **Text prompt input** describes the sound effect characteristics (type, duration, mood, context)
* **Duration parameter** controls the length of the generated sound effect in seconds
* Outputs audio in standard game-compatible formats for immediate engine integration
* Produces one-shot effects, ambient loops, UI sounds, and environmental audio from natural language descriptions
* Quality and acoustic properties vary based on prompt specificity and descriptive detail
* No audio source material required—generates entirely from text prompts

**Useful for:** Rapid prototyping of placeholder sound effects, generating UI audio feedback, creating one-off environmental sounds, iterating on audio design concepts during pre-production.

### Speech to Text

Converts spoken audio into text using selectable backend models. Accepts an audio input and produces a text transcription suitable for downstream processing in dialogue and workflow pipelines.

<figure><img src="/files/CcUjolWOVtmxJoPMCWnw" alt="" width="563"><figcaption></figcaption></figure>

* **Audio input** accepts recorded or generated speech for transcription
* **Text output** returns the transcribed content as a string
* Selectable backend models determine transcription accuracy and language coverage
* Output text can feed directly into dialogue nodes, subtitle generation, or text-processing workflows
* Suitable for batch transcription of voice recordings during development

**Useful for:** generating subtitles and captions for cutscenes, transcribing recorded voice lines into editable dialogue text, converting placeholder voiceover into script data, and building voice-command prototypes for in-game input systems.

## Common pitfalls

* **Using Simple Text to Speech for multi-character dialogue.** Simple Text to Speech performs single-voice synthesis and does not parse speaker labels, so a script with `A:` / `B:` prefixes is read as one voice. For distinct per-speaker voices, use Audio Speech Workflow or Text to Dialogue (ElevenLabs).
* **Skipping Voice Selection before bulk generation.** Voice Selection auditions voices from sample recordings without generating new audio; generating large amounts of dialogue first and discovering the default voice mismatches the character forces full regeneration.
* **Leaving duration unset on Sound Effects (ElevenLabs).** The duration parameter sets output length in seconds; without a value matched to the target context, generated effects run too short or too long for the intended in-game use.
* **Enabling strict duration when phrasing matters on Create Music (ElevenLabs).** Strict duration enforces exact length and can cut a track mid-phrase; leave it off when natural musical endings matter more than an exact length.
* **Expecting Simple Text to Speech to alter pitch or timbre.** It has no built-in voice modification. Chain it into Voice Changer (ElevenLabs) for pitch, formant, or timbre adjustments.

## Related nodes

* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — Input Audio supplies external audio for processing through Voice Changer (ElevenLabs) or Speech to Text. Input Text supplies dialogue scripts for text-to-speech nodes.
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — Text Generation (LLM), Combine Text, and Structured Output can generate or structure dialogue scripts that then feed audio nodes such as Text to Dialogue (ElevenLabs) and Simple Text to Speech.
* [Video Nodes](/atlas-ai-studio-overview/node-index/video-nodes) — Lipsync pairs audio generated by these nodes with a character image; Voice Changer (ElevenLabs) preserves the speech timing that Lipsync relies on.
* [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) — for exposing audio-generation workflows as callable endpoints for in-engine procedural audio and dynamic dialogue systems.

## Frequently asked questions

**What audio formats do the audio nodes produce?**

MP3 and WAV are standard outputs, compatible with common game engines for direct import.

**Can I use generated audio commercially in shipped games?**

Audio rights depend on the generation backend's terms (for example, ElevenLabs has its own commercial licensing). Verify the backend's terms before shipping generated audio. For risk-sensitive cases, treat generated audio as prototype-grade and replace it with licensed or studio-recorded audio before release.

**How do I generate dialogue with different voices for different characters?**

Use Audio Speech Workflow or Text to Dialogue (ElevenLabs) with speaker-labeled dialogue such as `A: Hello` and `B: Hi there`. The Add Audio Tags (ElevenLabs) node parses `A:` / `B:` / `C:` labels and assigns each unique speaker a distinct voice by label order, with no manual voice configuration. It supports unlimited speakers within a single dialogue block.

**How do I convert recorded voice lines into editable text?**

Use the Speech to Text node, which accepts recorded or generated speech and returns a transcription as a string. The output can feed dialogue nodes, subtitle and caption generation for cutscenes, or other text-processing workflows.

**Can the music generation nodes produce loopable tracks?**

Create Music (ElevenLabs) and Simple Music Generation can be prompted for loop-friendly compositions, though loop-point control is limited. For seamless looping, edit the output in a DAW or use shorter tracks designed to loop.

**What is the difference between Simple Text to Speech and Text to Speech (ElevenLabs)?**

Simple Text to Speech is a single-voice node with selectable backends for fast prototyping. Text to Speech (ElevenLabs) uses ElevenLabs voice models, with voice selection determining speaker identity, tone, and delivery style, and MP3 and WAV output options.

**How do I plan a complex multi-segment audio sequence before generating it?**

Use Composition Plan (ElevenLabs) to structure segment order, speaker assignments, and content direction, then feed that plan into Create Music (ElevenLabs), which accepts structured composition plan input for control over arrangement and timing. This lets you iterate on structure before synthesizing audio.

**How do I add sung vocals or remix an existing track?**

Use Simple Modify Music. Lyrics Only adds sung vocals over an instrumental from a text prompt, Inpainting regenerates a specified time range while preserving surrounding audio, Remix restructures tempo and arrangement, and Cover recreates the track with alternative instrumentation or vocals.

**Can I integrate audio nodes into a real-time game pipeline?**

Yes, via [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes). Export the audio workflow as an API and call it from your game backend or engine. Generation latency means on-demand audio suits non-time-critical content, such as post-action barks and ambient generation, rather than instant-response audio.


# API Nodes

API Nodes turn any Atlas workflow into a callable REST endpoint that runs from inside Unity, Unreal Engine, Blender, or any custom integration. Use them to deploy production-grade AI workflows directl

API Nodes are how Atlas workflows become production infrastructure. Build a workflow once in the Atlas visual editor, mark its inputs and outputs with API Nodes, and the workflow is instantly callable as a versioned REST endpoint. The same workflow now runs from Unity, Unreal Engine, Blender, an automation pipeline, or any custom backend that can make an HTTP request.

This page covers when to use API Nodes, the available input and output types, how to export a workflow as an API, common integration patterns, and the most frequent pitfalls when wiring Atlas into a production game pipeline.

## When to use API Nodes

Use API Nodes whenever the workflow needs to run outside the Atlas web UI:

* **Engine integration.** Trigger 3D asset generation from inside Unreal Engine or Unity. Players, designers, or game systems request assets on demand; Atlas executes the workflow and returns the result.
* **Batch processing.** Run the same workflow over hundreds or thousands of inputs (e.g., generating variants of a hero asset, processing a backlog of concept art). External scripts orchestrate the calls and collect results.
* **UGC pipelines.** Power player-generated content features where users enter prompts and the game returns custom 3D assets in real time.
* **Automation chains.** Wire Atlas into automation tools like n8n, Zapier, or custom CI/CD pipelines for asset processing as part of larger workflows.
* **Headless production environments.** Run Atlas workflows from Blender scripts, command-line tools, or server-side processes where opening the web UI is impractical.

If the workflow only runs interactively in the Atlas visual editor, API Nodes are not required. They become necessary the moment the workflow needs to execute from outside the editor.

There are two categories of API Nodes:

1. **Input API Nodes** — define what data your external request can send into the workflow.
2. **Output API Nodes** — define what the workflow returns after execution.

Once these nodes are placed, the workflow can be exported as an API directly from the toolbar.

### Input API Nodes <a href="#api-input-nodes" id="api-input-nodes"></a>

Input API Nodes replace native inputs so that external applications can provide data through the API.

#### Available Input Types <a href="#available-input-types" id="available-input-types"></a>

* **API Input Text** — accepts string data
* **API Input Number** — accepts numeric values
* **API Input Boolean** — accepts true/false flags
* **API Input Image** — accepts image file uploads
* **API Input Mesh** — accepts mesh files (GLB, OBJ)
* **API Input EXR** — accepts EXR files

Use these nodes when preparing your graph for external execution.

<figure><img src="/files/Xm4FLUtShqRLjtzshlWc" alt="" width="563"><figcaption></figcaption></figure>

### Output API Nodes <a href="#api-output-nodes" id="api-output-nodes"></a>

API Output Nodes define what the API returns when the workflow completes.

#### Available Output Types <a href="#available-output-types" id="available-output-types"></a>

* **API** **Output Text**– returns text strings
* **API** **Output Number**– returns numeric values
* **API** **Output Boolean**– returns true/false results
* **API Output Image** – returns generated images
* **API** **Output Mesh**– returns mesh files
* **API Output SVG** – returns vector graphics
* **API Output EXR** — returns EXR files

Replace the final node’s output with the corresponding API Output Node to expose your result.

<figure><img src="/files/Zx0g3qMULeHv5IwxtHLe" alt="" width="563"><figcaption></figcaption></figure>

### Exporting Your Custom Workflow as an API <a href="#exporting-your-custom-workflow-as-an-api" id="exporting-your-custom-workflow-as-an-api"></a>

<figure><img src="/files/HcIHFtNHBvv3kCtytyte" alt="" width="563"><figcaption></figcaption></figure>

After setting the inputs and outputs with API Nodes, you can export your workflow as an API:

#### Steps <a href="#steps" id="steps"></a>

1. Build your workflow as usual.
2. Replace **all required inputs** in your workflow with the corresponding **Input API Nodes**.
3. **Add** an Output API Node and connect it to your final output node to expose the result through the API.
4. Click the **Export API** icon in the toolbar.
5. Enter a name for your API.
6. Click **Create**.

<figure><img src="/files/aTJePLNrAkofJLvihTo1" alt="" width="563"><figcaption></figcaption></figure>

The platform then generates:

* **Bash example**
* **Python example**
* **Full API JSON specification**
* **Your API ID**

These can be used immediately to call your workflow programmatically.

<figure><img src="/files/HnJaTbg8LWQIPAs4nnAL" alt="" width="563"><figcaption></figcaption></figure>

### Using the Generated API <a href="#using-the-generated-api" id="using-the-generated-api"></a>

Once created, the API can be called from any environment:

* Python scripts
* Bash / curl
* Web backends
* Unity or Unreal Engine tools
* n8n or automation pipelines
* Blender scripts
* Custom UIs

Your API ID ensures that the workflow is executed with the exact graph and nodes you configured.

## Common integration patterns

A few patterns that tend to work well in production:

* **In-engine generation.** A game client (Unreal Engine, Unity) makes an HTTP request to the Atlas API endpoint, passes the player's prompt or reference image, waits for the workflow to complete, then loads the returned `.glb` directly into the scene. Best for slower, higher-quality generation moments. Pair with a loading state or async pattern in the engine to keep the experience smooth.
* **Background processing.** A server-side job queue (your own backend, or a tool like n8n or Temporal) calls Atlas in the background, stores results in a CDN or asset bucket, and serves them to clients on demand. Best for batch generation, UGC moderation pipelines, or pre-generation workflows.
* **Designer-facing tools.** A custom UI (web app, internal tool, or Blender plugin) wraps the API for designers, art directors, or technical artists to use without writing code. The tool sends parameters to Atlas, displays results inline, and gives designers a friction-light way to iterate.
* **Hybrid live and pre-generated.** Pre-generate a catalog of assets via batch API calls, then use live API calls only for player-specific or rare-case variants. Reduces latency and cost.

## Common pitfalls

Mistakes that consistently cause trouble when wiring up API Nodes:

1. **Forgetting to replace native inputs with API Input Nodes.** If a workflow still relies on the visual editor's interactive inputs (e.g., a text input typed directly into the node), the exported API has no way to receive that data. Every parameter the external caller needs to control must be wired to an API Input Node.
2. **Skipping the API Output Node.** Without an API Output Node attached to the final result, the API completes but returns nothing useful. Always connect the final node of your workflow to an API Output of the matching type (mesh, image, text, etc.).
3. **Hard-coding values that should be parameters.** If your workflow uses a fixed text prompt, fixed dimensions, or fixed model selection internally, the API will produce the same output every time. Promote anything that callers might want to vary into an API Input Node.
4. **Not versioning workflows before exporting.** Each export captures the workflow's exact graph state. If you later modify the workflow in the editor, the deployed API still runs the old version. Re-export to roll out changes, and consider keeping the old API ID alive during transitions so dependent clients have time to migrate.
5. **Treating the API as synchronous when it isn't.** Some Atlas workflows take 30 seconds or longer (especially 3D generation). Engine integrations should treat API calls as async by default, with proper loading states, timeout handling, and error fallback.

## Related nodes

* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — the in-editor counterparts to API Input Nodes
* [Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes) — workflows commonly exposed via API for image generation and processing
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — 3D generation and mesh processing workflows for engine integration
* [Utility Nodes](/atlas-ai-studio-overview/node-index/utility-nodes) — routing, conditionals, and orchestration patterns inside API workflows
* [3D Generation Best Practices](/atlas-ai-studio-overview/node-index/mesh-nodes/3d-generation-best-practices) — patterns for building robust 3D-generation APIs

## Frequently asked questions

**How does the Atlas API handle calls? Synchronous or asynchronous?**

The API is **async-only by design**. A workflow submission returns an `execution_id` immediately; you then poll the status endpoint (`/{version}/api_status/{execution_id}`) until completion, and download the output when ready. There is no blocking synchronous call. Plan your engine or backend integration around this pattern from the start.

**Can I call the same Atlas API from multiple engines and tools simultaneously?**

Yes. Once exported, the workflow is a standard REST endpoint and can be called from any HTTP-capable environment in parallel. The same workflow can serve Unity, Unreal, Blender, and a web backend at the same time.

**How long does an API call take?**

Latency depends on the workflow. Lightweight image edits return in a few seconds. Full 3D generation with PBR materials typically runs in 30 to 60 seconds. Workflows that chain many nodes are slower. The configurable client-side timeout (set on the Unity and Unreal plugins, or in your own HTTP client) should accommodate the slowest expected workflow plus margin.

**Can I update a workflow without breaking existing API callers?**

The API ID is tied to the workflow version at export time. Updating the workflow in the visual editor does not automatically update the deployed API; you need to re-export. This is intentional, so deployed integrations stay stable until you choose to roll out a new version. The URL pattern `{baseUrl}/{version}/api_execute_async/{apiId}` includes the version, so different versions can coexist.

**Does the API support file uploads?**

Yes, for mesh and image inputs via `multipart/form-data`. Upload a file to `/{version}/upload/{apiId}`, receive a `file_id` in the response, then reference that `file_id` in the workflow payload. Audio file inputs are not yet supported in this version.

**Can I run Atlas workflows inside Unreal Engine or Unity without writing my own HTTP client?**

Yes. Atlas ships official plugins:

* [Atlas Workflow Plugin for Unity](https://github.com/Atlas-Design/AtlasPlatform_UnityPlugin) — Unity 2023.1 or newer. Adds Atlas menus and editor windows that load a workflow, fill inputs, run, and import results directly into the project.
* [Atlas Workflow Plugin for Unreal Engine](https://github.com/Atlas-Design/AtlasPlatform_UnrealPlugin) — Unreal Engine 5.5 or newer. Provides Blueprint and C++ APIs for loading and executing workflows, including in packaged builds, not just the editor.

If you prefer to integrate directly without the plugins, the API is standard REST and works from any HTTP-capable language or environment.

**Where do I find the full API JSON specification?**

When you export a workflow as an API, the platform generates a downloadable JSON specification alongside the Bash and Python examples. The spec documents every input parameter, output type, and the request schema. Import it into Postman, Insomnia, or your OpenAPI tooling of choice for testing.

**What about authentication?**

Authentication is configured per workflow when you export the API. Credentials, the API ID, and the version are embedded in the workflow JSON file that the platform provides. Keep the workflow JSON file private if you're publishing your integration code (treat it as you would any credentials file).


# Utility Nodes

Utility nodes in Atlas AI Studio extract, transform, and recombine text, images, PDFs, and metadata for deterministic, reusable game asset workflows.

Utility Nodes provide general-purpose functions that support all Atlas workflows.\
They are not tied to image or mesh generation; instead, they handle documents, text processing, arrays, and lightweight LLM operations.

Utility Nodes are essential for orchestrating complex workflows, preparing inputs, and structuring data for generation nodes.

## When to use utility nodes

Use utility nodes to glue the rest of your workflow together. The most common patterns:

* **Document-driven content generation.** Use Extract Text From Document and Extract Images From Document to pull material out of a GDD or art bible, then feed the extracted content into generation nodes via Combine Text or directly.
* **Multi-step LLM reasoning.** Chain Text Generation (LLM) nodes with Combine Text to build conditional prompts where one step's output becomes the next step's instruction. Powerful for "describe → reformat → generate" pipelines.
* **Batch operations.** Use Create Images List, Create Text List, and Create Number List to build arrays that downstream generation nodes process in parallel, then Break \* List to route outputs to specific branches.
* **Character sheet workflows.** Pair Split Character Sheet with Multi-View to 3D for converting a multi-angle drawing into a rigged 3D character in one workflow.
* **Structured output for game logic.** Use Structured Output Node to extract typed JSON (NPC stats, loot tables, dialogue metadata) from LLM responses, ready for direct consumption by game systems.

## Document Nodes <a href="#document-nodes" id="document-nodes"></a>

### Extract Images From Document <a href="#extract-document-images" id="extract-document-images"></a>

This node processes an uploaded **PDF document** and extracts all images embedded within it.

* Input: **PDF file** (e.g., a Game Design Document)
* Output: **Image Array**
* Each extracted image is returned as an element in the array.

<figure><img src="/files/epe3mb79u0Pdv10xoGBP" alt="" width="563"><figcaption></figcaption></figure>

You can use the resulting array with:

* **Find Images by Description** node to pull specific images

<figure><img src="/files/d9GVV6xxJ3fJBYiJpOnI" alt="" width="563"><figcaption></figcaption></figure>

* **Image Generation nodes** that accept image arrays
* **Break Images Array** to isolate specific images

### Extract Text From Document <a href="#extract-document-text" id="extract-document-text"></a>

Extracts text content from a **PDF file** and separates it into three categories:

* **Visual Descriptions** — descriptions of scenes, objects, characters
* **Relevant Other Text** — supporting information that may assist generation
* **Filtered / Irrelevant Text** — removed noise, metadata, or non-useful content

<figure><img src="/files/4lFHnz5Qic4kBjf2v7X3" alt="" width="463"><figcaption></figcaption></figure>

This separation is useful when you want to feed only contextually relevant text into your generation nodes or LLM prompts.

### Split Character Sheet

Detects and extracts individual views—such as front, side, and back—from a single character design sheet.

* Automatically parses a design sheet to isolate different character poses into separate image outputs.
* Works best with horizontal, vertical, or grid layouts where poses are clearly separated by empty space.
* Outputs: Specific views for Front, Right, Left, and Back, plus an array of any "Other" detected angles (like 3/4 views).
* Ideal for: Preparing character sheets for the Multi-View -> 3D node or organizing concept art.

<figure><img src="/files/W8UrF9v3YZE8Ok61cKUZ" alt="" width="563"><figcaption></figcaption></figure>

## Text & LLM Nodes <a href="#text--llm-nodes" id="text--llm-nodes"></a>

### Combine Text <a href="#text-concatenate" id="text-concatenate"></a>

Merges multiple text inputs into one unified string.

* Helps combine text blocks before sending them into generation nodes
* Useful when you want to enforce a **fixed prefix or order**\
  (e.g., system constraints + extracted text → final generation prompt)

Typical use case:

* Merge a “generation instruction” with extracted document text
* Feed the combined result into an image or 3D generation node

<figure><img src="/files/rtgIEIeIiY3CMUNpaEJG" alt="" width="563"><figcaption></figcaption></figure>

### Text Generation (LLM) <a href="#simple-llm-call" id="simple-llm-call"></a>

LLM tool for controlled text generation.

* Inputs:
  * **System Prompt** — defines the agent’s role and behavior
  * **Text Input** — any context or content to transform
* Output: **Single text result**

<figure><img src="/files/IiWFycXsW3cYJSlaKKVI" alt="" width="563"><figcaption></figcaption></figure>

Example use case:

You can input a concept art image and use **Describe Image** to extract the existing character list. A Text Generation (LLM) can then be instructed to return only the first character from that list. You may add an additional text input that defines how the multimodal node should isolate the selected character for 3D modeling. Using Combine Text, you merge the LLM-generated character description with your isolation instruction and feed the combined text directly into the multimodal node. This creates a controlled, image-conditioned prompt for generating a clean extraction of a single character ready for modeling.

Another example is providing an image of a full scene and using **Describe Image** to obtain a structured description of its spatial layout. You can then instruct the **Text Generation (LLM)** to act as a spatial-logic prompt generator that transforms these descriptions into a precise prompt for generating a 2D top-down plan of the same scene. By passing this prompt into a multimodal node, you obtain a clean plan abstraction. With this method, any input image produces a custom, image-specific plan prompt through the combined use of Describe Image, LLM processing, and Combine Text.

### Simple Text Render

Generates a 2D image from a text string with automatic height adjustment to fit the content.

* Allows for basic typography control, including font selection, size, line spacing, and color customization.
* The output image width is fixed by the user, while the height scales dynamically based on how much text is provided.
* Inputs: Input Text, Font Name (Arial/Nunito), Image Width, Font Size, and Background/Font Colors.
* Ideal for: Creating labels, UI elements, or adding textual descriptions directly into your image processing pipeline.

<figure><img src="/files/wX8nwRPcP75vQ33ynG5p" alt="" width="563"><figcaption></figcaption></figure>

## Array Management Nodes <a href="#array-management-nodes" id="array-management-nodes"></a>

### Concatenate Images

Combines multiple images into a single image file by joining them side-by-side or top-to-bottom.

* In Horizontal mode, all images are automatically scaled to match the height of the first image in the array.
* In Vertical mode, all images are scaled to match the width of the first image in the array to maintain a uniform column.
* Inputs: Input Images (Array), Direction (Horizontal/Vertical).
* Ideal for: Creating comparison grids, before-and-after montages, or reassembling split character sheets

<figure><img src="/files/EVHu38XNbJH6OhzIkmI4" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/wdrZYIpJkQYlTa900OwG" alt="" width="563"><figcaption></figcaption></figure>

Utility Nodes allow these types of modular logic chains ; extracting structured information, refining or transforming it, and recombining it to produce workflows that adapt automatically to the input image while remaining deterministic and reusable across the Atlas platform.

### Structured Output Node

Generates JSON responses constrained to a user-defined schema, ensuring validated, parseable output for programmatic use. Small models are blocked from complex schemas to maintain reliability.

* Accepts a JSON schema definition alongside a text prompt; the output strictly conforms to the schema structure.
* Returns validated JSON suitable for direct parsing into game logic, asset metadata, or procedural generation parameters.
* Schema enforcement prevents hallucinated fields or malformed data structures.
* Inputs: Prompt (String), JSON Schema (Object), Model (Selector).
* Outputs: Structured JSON (Object).
* Useful for: Extracting NPC attributes from narrative descriptions, generating itemized loot tables, parsing scene composition data, or creating dialogue trees with typed metadata fields.

<figure><img src="/files/A1WQGw7osD1BEwRojoSQ" alt="" width="563"><figcaption></figcaption></figure>

### Extract PNG Metadata

Reads generation parameters, prompts, and model settings embedded in PNG files produced by image generation nodes, enabling prompt reuse and parameter inspection across workflows.

* Extracts embedded metadata including the original prompt, seed, model identifiers, and generation parameters stored in PNG comment chunks.
* Returns structured data fields that can be passed directly to other nodes for iterative refinement or batch reproduction.
* Works with PNG files generated within Atlas or imported from compatible external tools that follow standard metadata conventions.
* Displays specific error messages when metadata is missing or malformed, rather than silently failing.
* Inputs: PNG Image (File).
* Outputs: Prompt (String), Seed (Integer), Model (String), Parameters (Object).
* Useful for: Reverse-engineering successful asset generation settings, building remix pipelines that preserve original prompts, debugging generation parameters, or creating variant workflows that iterate on existing images while maintaining audit trails.

<figure><img src="/files/fkGE6BkgTFXOppswRpxG" alt="" width="563"><figcaption></figcaption></figure>

### Create PDF List

Combines multiple PDF inputs into a single ordered array for batch processing or sequential analysis workflows.

* Accepts individual PDF file inputs and merges them into a unified list structure.
* Preserves input order, enabling deterministic downstream operations on multi-document sets.
* Outputs: PDF List (Array).
* Inputs: PDF 1, PDF 2, PDF 3... (variable count, PDF files).
* Useful for: Aggregating quest documentation before extraction, batching asset reference sheets for analysis, or organizing multi-chapter narrative PDFs for sequential dialogue parsing.

<figure><img src="/files/DpZgFhC0t2Phv0hseNUU" alt="" width="563"><figcaption></figcaption></figure>

### Create Text List

Assembles multiple text inputs into a single ordered list, enabling batch operations and multi-prompt workflows.

* Accepts up to eight individual text inputs (Text 1 through Text 8); empty inputs are ignored.
* Outputs a unified text array preserving input order for downstream nodes that process lists.
* Inputs: Text 1–8 (String, optional).
* Outputs: Text List (Array).
* Enables parallel processing: combine asset names extracted from concept art with style descriptions to generate variations across multiple objects in one pass.
* Useful for: Batch-generating sprite variants, assembling dialogue options for NPC responses, feeding multiple prompts into image generation arrays, or combining extracted metadata with templated instructions.

<figure><img src="/files/C0EMiZgc1l1XIJ85ETGe" alt="" width="563"><figcaption></figcaption></figure>

### Break Text List

Splits a concatenated text list back into individual text outputs, reversing the operation performed by the Concatenate Text node. Each line or delimiter-separated segment becomes a discrete output port.

* Accepts a single combined text input and separates it based on newline or custom delimiters.
* Outputs multiple individual text strings, one per parsed segment.
* Output port count matches the number of segments detected in the input list.
* Inputs: Combined Text (String), Delimiter (String, optional).
* Outputs: Text outputs (dynamic count based on input).
* Preserves original segment order and whitespace handling per delimiter configuration.
* Useful for: Unpacking batch-generated NPC names or dialogue lines, distributing localized strings to separate UI elements, splitting procedurally generated item descriptions, or routing individual quest objectives to distinct logic branches.

<figure><img src="/files/d4C8IvXFJJQKevv9KP3s" alt="" width="563"><figcaption></figcaption></figure>

### Break PDF List

Splits an array of PDF files into individual PDF outputs, enabling parallel processing or selective routing of multi-document collections.

* Accepts a PDF List (Array) input and exposes each element as a separate PDF output port.
* The number of output ports matches the array length; each port emits one PDF document in sequence.
* Enables conditional logic, per-document transformation, or targeted extraction when combined with downstream nodes like Extract PDF Text or Convert PDF to Images.
* Inputs: PDF List (Array).
* Outputs: Individual PDF files, one per array element.
* Useful for: Batch-processing narrative design documents, splitting multi-chapter story bibles for parallel text extraction, or isolating concept art pages for individual image analysis and asset tagging.

<figure><img src="/files/OslZo7yB54xvQA7vdn3C" alt="" width="563"><figcaption></figcaption></figure>

### Create Number List

Combines individual number inputs into a single array for batch processing or iteration in downstream nodes.

* Accepts up to eight separate number inputs (Number 1 through Number 8).
* Outputs a unified number array containing all provided values in order.
* Empty or unconnected inputs are omitted from the resulting list.
* Inputs: Number 1–8 (Float/Integer).
* Outputs: Number List (Array).
* Useful for: Feeding parameter sets to image generators, defining animation keyframe timings, batching seed values for asset variations, or programmatically controlling multiple effect intensities in a single pipeline.

<figure><img src="/files/SNJZxiJ38QwpHljhbplf" alt="" width="563"><figcaption></figcaption></figure>

### Create Image List

Combines multiple individual image inputs into a single ordered array, enabling batch operations and multi-image workflows. This node consolidates separate image connections into one unified list that downstream nodes can iterate over or process collectively.

* Accepts any number of individual image inputs and outputs them as a single Image Array.
* Input images retain their original resolution and format; no scaling or transformation is applied.
* Output order matches the sequence of connected inputs from top to bottom.
* Inputs: Image 1, Image 2, Image 3... (expandable).
* Outputs: Image List (Array).
* Essential for feeding multiple texture variations into batch processors, assembling reference sheets for style transfer, or preparing image sets for concatenation and comparison grids.

<figure><img src="/files/YZ80Kte40hdyXjj33QXK" alt="" width="563"><figcaption></figcaption></figure>

### Concatenate Image Arrays

Merges multiple image array inputs into a single unified array, preserving order and enabling batch operations across collections from different sources.

* Accepts two or more image array inputs and outputs a single flattened array containing all images in sequence.
* Order is determined by input port order; images from the first connected array appear before those from subsequent arrays.
* Does not modify, resize, or composite the images themselves—purely a collection merge operation.
* Inputs: Image Array 1 (Array), Image Array 2 (Array), \[additional array ports as needed].
* Outputs: Merged Images (Array).
* Useful for: Combining character variations from multiple generation passes, pooling environment assets before batch processing, assembling multi-source texture sets for style transfer, or consolidating prototype sprites for export.

<figure><img src="/files/xBiVZn5UxybzsIaU1jvw" alt="" width="563"><figcaption></figcaption></figure>

### Break Image List

Unpacks an image array into separate, individually accessible image outputs for downstream nodes that expect single images rather than collections.

* Accepts an image array input and exposes each element as a discrete numbered output port (Image 1, Image 2, etc.).
* Maximum output count matches the array length; empty slots remain inactive if the array is shorter than the node's port capacity.
* Enables routing different images from a batch to different processing branches without iteration logic.
* Inputs: Image List (Array).
* Outputs: Image 1, Image 2, Image 3... (Individual Images).
* Useful for: Applying distinct post-processing filters to each generated character variation, routing specific frames from a multi-frame generation to separate composition nodes, or feeding individual concept sketches into parallel refinement workflows.

<figure><img src="/files/QBTtZly9RxS8UiZpovwY" alt="" width="563"><figcaption></figcaption></figure>

### EXR to 16-bit PNG

Converts an EXR image into a 16-bit PNG, preserving the extended tonal range of the source across a widely compatible file format.

<figure><img src="/files/rAdRiZl3Kwa3oKgbUT9S" alt="" width="563"><figcaption></figcaption></figure>

* **Input.** A single EXR file.
* **Output.** A 16-bit PNG containing the converted image data.
* **Bit depth.** Writes 16 bits per channel, retaining finer gradients than 8-bit output.
* **Format bridging.** Moves high-dynamic-range renders into a format accepted by common engine and DCC import pipelines.
* **Typical use.** Prepare rendered height maps, normal data, or lighting passes for engines that do not ingest EXR directly.

Useful for: converting HDR render outputs into engine-ready textures for terrain, lighting, and material workflows.

**Example:** Example use case for the EXR to 16-bit PNG node.

<figure><img src="/files/2BUZMDzIuvgv0B2mctyv" alt="" width="563"><figcaption></figcaption></figure>

## Common pitfalls

* **Using a Text Generation (LLM) call where string concatenation suffices.** When a workflow only needs to join text with Combine Text, routing that text through Text Generation (LLM) introduces model variability into a step that could be deterministic, producing inconsistent output across runs.
* **Feeding shorter-than-expected arrays into Break Image List.** Its output count matches the array length, so ports beyond the array size stay inactive. Downstream branches wired to those ports receive nothing and sit idle without an obvious error.
* **Concatenating images with mismatched aspect ratios.** Concatenate Images scales inputs to the first image's height in Horizontal mode and to its width in Vertical mode. Inputs with very different proportions come out stretched or squashed; normalize sizes upstream when uniformity matters.
* **Expecting Split Character Sheet to parse any layout.** It relies on empty space separating poses in horizontal, vertical, or grid arrangements. Overlapping or tightly packed views fail to isolate cleanly into the Front, Right, Left, Back, and Other outputs.

## Related nodes

* [Input Nodes](/atlas-ai-studio-overview/node-index/input-nodes) — the source nodes whose uploaded PDF and image content utility nodes transform.
* [Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes) — generation and editing targets for utility-prepared image arrays and concatenated grids.
* [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes) — Split Character Sheet isolates individual character views for the Multi-View to 3D node.
* [API Nodes](/atlas-ai-studio-overview/node-index/api-nodes) — for exposing utility-orchestrated workflows as external APIs.

Utility nodes form modular chains that extract structured information, transform it, and recombine it. Nodes such as Extract Text From Document, Text Generation (LLM), Combine Text, and Structured Output Node connect the input, image, and mesh stages listed above.

## Frequently asked questions

**Can I use utility nodes without any generation nodes in the workflow?**

Yes. A workflow can be built entirely from utility nodes: extract text from a PDF with Extract Text From Document, transform it with Text Generation (LLM), and emit validated JSON with the Structured Output Node. No image or mesh generation is required, which makes utility-only chains suitable for document-processing pipelines and metadata extraction.

**How do I combine outputs from multiple text sources into one prompt?**

Use Combine Text to merge multiple text inputs into a single string. The input order determines the order in the final string, so place fixed instructions or system constraints first and dynamic content, such as extracted document text, afterward. This is a common way to enforce a fixed prefix before feeding a prompt into an image or 3D generation node.

**Can the Structured Output Node guarantee schema compliance?**

The Structured Output Node enforces the user-defined JSON schema on the model output and returns validated JSON that parses directly into game logic, asset metadata, or procedural generation parameters. Schema enforcement prevents hallucinated fields and malformed structures. Small models are blocked from complex schemas, and the node returns an error rather than malformed JSON when the model cannot satisfy the schema. For production-critical output, use larger models and design schemas conservatively.

**How do I split a batch of images or text back into individual items?**

Use Break Image List to unpack an image array into separate numbered image outputs, one port per array element, so different images can route to different processing branches. Use Break Text List to split a combined text input into discrete outputs by newline or a custom delimiter, preserving segment order. Break PDF List performs the equivalent operation for a PDF array.

**How do I assemble multiple inputs into one array for batch processing?**

Use Create Text List (up to eight text inputs), Create Number List (up to eight number inputs), Create Image List (expandable image inputs), or Create PDF List to build ordered arrays. Empty or unconnected inputs are omitted from text, number, and PDF lists. Concatenate Image Arrays merges two or more existing image arrays into one flattened array in input-port order without resizing or compositing.

**Can I reuse the generation settings stored in a PNG?**

Yes. Extract PNG Metadata reads the original prompt, seed, model identifier, and generation parameters embedded in PNG comment chunks by image generation nodes. The returned fields can be passed to other nodes for iterative refinement or batch reproduction. It works with PNGs generated within Atlas or imported from tools that follow standard metadata conventions.

**How do I create a text-to-video or lipsync node from utility nodes?**

Utility nodes do not perform video or lipsync generation themselves; they prepare and route content for the generation nodes documented in [Image Nodes](/atlas-ai-studio-overview/node-index/image-nodes) and [Mesh Nodes](/atlas-ai-studio-overview/node-index/mesh-nodes). Utility nodes handle the surrounding steps such as text merging, list assembly, and metadata extraction.

**How do I move an EXR render into an engine that does not accept EXR?**

Use EXR to 16-bit PNG to convert an EXR file into a 16-bit-per-channel PNG, preserving finer tonal gradients than 8-bit output. This bridges high-dynamic-range renders such as height maps, normal data, and lighting passes into a format accepted by common engine and content-creation import pipelines.


# Release Notes


# Release 0.6.0

Platform link: [https://app-tech.atlas.design](https://app-tech.atlas.design/)

* Global node outputs cache:
  * Basically if you generate “a cat” with seed “42” you will have instant output, as it will now just retrieve it from the global cache ! (ofc. any change to inputs will not retrieve anything, but generate from scratch)
* API export changes (note that API export feature still needs to be enabled per-user and is disabled by default)
  * Executing API won’t recalculate nodes not dependent on the API inputs, and will use the global cache feature – so calling API with same inputs twice will result in near instant results !
  * API export feature now outputs “API info” JSON, which can be pasted directly to the Atlas Unreal plugin
  * **IMPORTANT: no backwards compatibility of already exported APIs with previous version, you need to re-export your APIs from app-tech.atlas.design (though there is backwards compatibility for projects, only not for already exported APIs) – we also plan to maintain backwards compatibility from now on**
* Enable bunch of new nodes for managing image arrays
  * Create and break image array
  * concatenate image arrays
  * retrieve specific images count from array by text prompt
  * “Multimodal” node with array input
  * image to text with array input
* Support for PDF documents input
  * drag & drop a pdf
  * use “Extract Document Images” and “Extract Document Text” nodes
    * Extract Document Images experimentally supports (automatically) extracting separate images from raster collages
* New “Simple LLM call” node – you can now use LLMs inside our app and exported APIs. One could even use our app to build just LLM pipelines without touching images or meshes.
* Various UI improvements:
  * We began a major refactoring of the UI to unify styles and behaviors across various pages and components created at different times with differing visions. Our goal is to create a solid and consistent platform UI.
  * After modifying hotkeys, you may need to refresh the page for the changes to take effect.
* Make the suggestions system more concise about amount of relevant suggestions
* Numerous bugfixes

### Known issues: <a href="#known-issues" id="known-issues"></a>

* After typing a prompt, you need to click out of the text field for the run button to be clickable/enabled
* “Selection agent” panel often jumps to the center of the screen
* some PDF documents may fail image or text extraction
* global cache via API doesn’t work when API has any binary inputs (images or meshes), will be fixed in next release/patch


# Release 0.7.0

Platform link: <https://app-tech.atlas.design/>

## API export changes: <a href="#api-export-changes" id="api-export-changes"></a>

* simple workflows with images will work now much faster as we skip unnecessary image captioning logic
* nodes which are not preceding output nodes are automatically removed from exported graph (with a warning message in the API creation UI)
* Fixed global cache with workflows reusing the same input file id (it didn’t cache previously always rerunning whole pipeline part dependent on the input images or meshes)
* API export/execution is backwards compatible with APIs exported using 0.6.0 release
* API export dialog improvements:
  * copy-able python code along bash and API info
  * button for only copying API id for fast replacement
  * fixed bug with not copy-able text on this dialog
  * fixed issue with api name not properly displaying in copied API info

## Optimization: <a href="#optimization" id="optimization"></a>

* Nodes which output images display results now 8-10 seconds faster, because now the new image is being captioned (for suggestions system) in the background.

## Text to origin <a href="#text-to-origin" id="text-to-origin"></a>

<figure><img src="/files/hYxDvPRtKmoEXCG3jAou" alt="" width="563"><figcaption></figcaption></figure>

New experimental node, allowing to set 3D origin in a desired location. It’s still a fresh feature and needs some improvements as failure cases are often.

## New models/algorithms: <a href="#new-models-algorithms" id="new-models-algorithms"></a>

* “Tile 4x” backend for upscale node - good for photos, upscales 4x without limitation, e.g. image can be 2048x2048 → will be 8192x8192. Not recommended for super high resolutions, may have some tiling artifacts
* “CGI” backend for upscale node - good for cgi, digital art, render images. Not good for photos.
  * `Maximum upscale for CGI is 3840x2160 (not applicable to a 9:16, e.g. 9:16 will be 1680x2160)`

## Automatic mesh frontalization: <a href="#automatic-mesh-frontalization" id="automatic-mesh-frontalization"></a>

<figure><img src="/files/MUGoTOqf14Q0GZRddzXT" alt="" width="480"><figcaption></figcaption></figure>

Auto transform has now 3 input booleans, previously it behaved like with only first 2 enabled always.

## Simple node arrangement: <a href="#simple-node-arrangement" id="simple-node-arrangement"></a>

* Add a simple automatic node placement cleanup, rearranging nodes to remove overlaps between them. There are 3 buttons, each is corresponding to different “strength”/ how much nodes are being moved from their initial positions.

  <figure><img src="/files/SwIPyJzYIM5Yfw4ivr9U" alt="" width="156"><figcaption></figcaption></figure>

## Other changes/fixes <a href="#other-changes-fixes" id="other-changes-fixes"></a>

* fixed project “Save and Edit” feature (issue with copied project being empty)
* added “other irrelevant text” output (as a 3rd one) for Document text extraction node
* added new simple “Math Expression” node
* fixed issue with pasting nodes pasting duplicates
* fixed issue about run button for node with required text input not updating after just typing text (required defocus)
* improved the node run button locked behavior when typing prompt
* fixed copy-paste issues

## UI changes: <a href="#ui-changes" id="ui-changes"></a>

* Nodes errors now properly display in frontend, so there shouldn’t be any “mysterious” cases of nodes just stopping running
* Projects view redesign
* vertical/horizontal UI switch
* node layout improvements


# Release 0.8.0

## New nodes: <a href="#new-nodes" id="new-nodes"></a>

* outpainting node:

<figure><img src="/files/YnAsQf31JeKEBzLXp3FD" alt="" width="563"><figcaption></figcaption></figure>

* remove background node

<figure><img src="/files/pKj4NRSnwAP1o5xttKro" alt="" width="563"><figcaption></figcaption></figure>

* video generation nodes

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fa8QzodwLT4IdPJ43pnUq%2Fvideo_1280.mp4?alt=media&token=bc54588e-7198-4bbc-bbb1-c46d56718f94>" %}

## Existing node improvements <a href="#existing-node-improvements" id="existing-node-improvements"></a>

* improved “Text to Origin” node reliability
* new “Reve” image generation backend(s)

<figure><img src="/files/EIXPNPZZFLgmO8OX6stn" alt="" width="563"><figcaption></figcaption></figure>

## UI <a href="#ui" id="ui"></a>

* automatically fit the viewed 3d object in the web 3d viewer (default binding to re-center again is “f” key).

## API export <a href="#api-export" id="api-export"></a>

* Added 2 options for auto generated python code examples

<figure><img src="/files/xFwlZNqf4nEZT0GlwtDs" alt="" width="563"><figcaption></figcaption></figure>

## Fixes <a href="#fixes" id="fixes"></a>

* fixed numerous “Simple 3D Scene” node internal failure cases
* fix browser sessions disconnecting and requiring page refresh after some idle time
* fix 3d mesh preview for meshes without UVs


# Release 0.9.0

Platform link (same as usual): [https://app-tech.atlas.design](https://app-tech.atlas.design/)

## UI <a href="#ui" id="ui"></a>

* UI was completely revamped from login screen, projects view and graph view

## New image to 3D models <a href="#new-image-to-3d-models" id="new-image-to-3d-models"></a>

* Hunyuan 3D 3.0 (check out it’s low poly mode!)
* Hitem 3d

<figure><img src="/files/GReIqmyyoBiNkstyQOlZ" alt="" width="563"><figcaption></figcaption></figure>

## New image generation/edition models <a href="#new-image-generation-edition-models" id="new-image-generation-edition-models"></a>

* Gemini 3 Pro image

<figure><img src="/files/GReIqmyyoBiNkstyQOlZ" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/8ERqiKENtcDF1BMwINaC" alt="" width="563"><figcaption></figcaption></figure>

* camera angle

<figure><img src="/files/VbytgxJlikVK3eZycI2x" alt="" width="563"><figcaption></figcaption></figure>

## Remesh <a href="#remesh" id="remesh"></a>

* now the “Remesh” node supports preserving the input mesh texture

## LLMs <a href="#llms" id="llms"></a>

* updated supported models by SimpleLLM node to support newest released LLMs of OpenAI and Google
* added same LLMs for “Make simple 3D Scene” node

## Fixes and optimizations <a href="#fixes-and-optimizations" id="fixes-and-optimizations"></a>

* suggestion engines should work faster and are more helpful with image arrays
* uploading a lot of files is much more reliable now


# Release 0.10.0

Platform link (same as usual): [https://app-tech.atlas.design](https://app-tech.atlas.design/)

## New models <a href="#new-models" id="new-models"></a>

* Added claude opus 4.5 support in “Simple llm call” and “make simple 3d scene” nodes

<figure><img src="/files/xsa8KN5WamzrDvrAk1vl" alt=""><figcaption></figcaption></figure>

* New image upscale backend (gemini image 3), and make it default

<figure><img src="/files/ffLBp0sHbZglqiKt8MlD" alt=""><figcaption></figcaption></figure>

* Added new Flux 2 family models backends for image generation/edition nodes

<figure><img src="/files/i1FxGYgv8O2R0bYgEYrS" alt=""><figcaption></figcaption></figure>

* Added new “Texture Mesh” node backend – Meshy (might be unstable for meshes > 35 mb)

<figure><img src="/files/S4Wy1uxeiNr6NESdsk0f" alt=""><figcaption></figcaption></figure>

* Added gemini-2.5-flash support for quality text to image node

## UI changes <a href="#ui-changes" id="ui-changes"></a>

* Added “project not found” text on invalid URLs (e.g. when projects not properly shared via share link), instead of blank project view
* Updated quality text to image and precise text edit nodes help strings, modify defaults, and reorder the backends list to be more reasonable
* Added save to file buttons on API export dialog:

  <figure><img src="/files/pBbkultOoU63WZU3vLyV" alt="" width="563"><figcaption></figcaption></figure>

## 3D transform nodes <a href="#id-3d-transform-nodes" id="id-3d-transform-nodes"></a>

* Added preserve aspect ratio param to the simple bbox fit node
* added thin geometry tolerance to the simple bbox fit node
* added “mass center” options for the origin set node

<div align="center"><figure><img src="/files/HdeGIExrAtz8OGxUaEJA" alt="" width="563"><figcaption></figcaption></figure></div>

With this workflow, each helmet will be “aligned” in the same way, ignoring thin geometry details which “stick out”You can see that helmets are alignedsimilar can be attempted with various sofas generation (aligned, ignoring “sticking out” geometry)

<figure><img src="/files/TQZ92qnSJgS81ftBhCIK" alt="" width="563"><figcaption><p>You can see that helmets are aligned</p></figcaption></figure>

<figure><img src="/files/aErRuiI0v0i0Y9L0s76Y" alt="" width="563"><figcaption><p>similar can be attempted with various sofas generation (aligned, ignoring “sticking out” geometry)</p></figcaption></figure>

## Fixes <a href="#fixes" id="fixes"></a>

* Fixed top left corner node menu not working (with help information)

<figure><img src="/files/99anGlMdALkd89wJ5FXC" alt="" width="480"><figcaption></figcaption></figure>


# Release 0.14.0

## NOTE <a href="#note" id="note"></a>

This release note includes all changes since 0.10.0 ! (we didn’t write release notes since then despite continuing releasing, so here goes a longer one).

**Release Date:** January 2026

**Previous Version:** v0.10.0 (December 1, 2025)

This release includes 137 commits with major improvements across nodes, AI assistant capabilities, UI/UX, pricing infrastructure, and backend architecture.

***

### 1. New Nodes & Backends <a href="#id-1.-new-nodes-and-backends" id="id-1.-new-nodes-and-backends"></a>

#### 3D Generation & Mesh Processing <a href="#id-3d-generation-and-mesh-processing" id="id-3d-generation-and-mesh-processing"></a>

* **Multi-view to 3D Node** – Converts multi-view images into 3D meshes with support for multiple backends

<figure><img src="/files/sSeN7Y0xgjHUHdyRLvBo" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/Oeiu5K0c26JU3iO9Vsnu" alt="" width="563"><figcaption></figcaption></figure>

* **Fast Image to 3D Node** – Quick image-to-3D conversion for rapid prototyping

<figure><img src="/files/7qcJwVSAgO3pNz8lJc3d" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/pXuEodhiDxXWwLSsb8hK" alt="" width="563"><figcaption></figcaption></figure>

* **Depth Estimation Node** – Generates depth maps from images

<figure><img src="/files/VbcHUGGR3KoQTGcO6MU6" alt="" width="563"><figcaption></figcaption></figure>

* **Mesh Texture Extraction & Application Nodes** – Extract textures from meshes and reapply them

<figure><img src="/files/xZddwIWMV0hStdjrRXEf" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/ZOT8QG8ceqpz2PSQsodt" alt="" width="563"><figcaption></figcaption></figure>

* **Remesh Node** – Refactored to support multiple backends (Hunyuan, Trellis2, etc.)

<figure><img src="/files/JoZEAbWRTtbzTKOG8GuX" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/moxUa3MviadhXsbdJlJB" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/Bw36bR7JnG5fQlLwFRcj" alt="" width="563"><figcaption></figcaption></figure>

* **Preview Mesh Compression** – Optimized mesh preview delivery
* **Hunyuan Parts Node** – Separates 3D object parts using Hunyuan

<figure><img src="/files/MhLBkSHRpQq7u6JrnsHm" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/cqg4pfDfaPaIKcArZE49" alt="" width="408"><figcaption></figcaption></figure>

* **Meshy Lowpoly Backend** – Added support for low-poly mesh generation

<figure><img src="/files/IXU1zQde4rKQuf4iIcNO" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/NYW839YYkR03rb0fBeBz" alt="" width="563"><figcaption></figcaption></figure>

#### Experimental Humanoid Animation Pipeline <a href="#experimental-humanoid-animation-pipeline" id="experimental-humanoid-animation-pipeline"></a>

* **Rig Humanoid Mesh Node** – Automatic rigging for humanoid 3D models

<figure><img src="/files/OmdDg9MLbO7ndIjgQ0Fv" alt="" width="563"><figcaption></figcaption></figure>

* **Animate Rigged Mesh Node** – Apply animations to rigged humanoid meshes

<figure><img src="/files/HHnlcDj013VDBhuP6FPT" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/tt5DzHj8ldwKbZ3pAzGq" alt="" width="563"><figcaption></figcaption></figure>

* **Split Character Sheet Node** – Splits sprite/character sheets into individual frames

<figure><img src="/files/fTscRCiAq9UxLwi4t4nm" alt="" width="563"><figcaption></figcaption></figure>

#### Image Generation Models <a href="#image-generation-models" id="image-generation-models"></a>

* **FLUX.2 Family** – Added FLUX.2 Max, Turbo, and Flash models

<figure><img src="/files/QuB9K3sKAqlYY45TIq7p" alt="" width="563"><figcaption></figcaption></figure>

* **ImagineArt 1.5 Pro Preview** – New image generation backend

<figure><img src="/files/2Z81Ca3j59TFiEEhNpTU" alt="" width="563"><figcaption></figcaption></figure>

* **Z-Image & P-Image Family** – Integrated additional image generation models

<figure><img src="/files/SqUvqgz7WmD3wrJGuv4Y" alt="" width="563"><figcaption></figcaption></figure>

* **Kling O1 Image** – Added Kling O1 image generation support

<figure><img src="/files/un1VADxZqcyv5dRDWIwi" alt="" width="563"><figcaption></figcaption></figure>

* **GPT-Image 1.5** – Upgraded with precise pricing support
* **Hunyuan Image 3.0 Instruct** – Added Hunyuan Image 3.0 image generation support

<figure><img src="/files/znJEfSx3rQmQjdP029Oe" alt="" width="563"><figcaption></figcaption></figure>

* **Grok Imagine** – Added Grok Imagine image generation support

<figure><img src="/files/9yE8bEpN22nBmjrsmiFl" alt="" width="563"><figcaption></figcaption></figure>

* **Gemini Upscale Backend** – Added preset with heavy image restoration

<figure><img src="/files/dtd8T24P8XzFRIrgS5Tf" alt="" width="563"><figcaption></figcaption></figure>

#### 3D Model Backends <a href="#id-3d-model-backends" id="id-3d-model-backends"></a>

* **Hunyuan3D v3 & 3.1** – FAL backend with Tencent fallback

<figure><img src="/files/fH9qvKuxqqIEoShnknlR" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/JTgFoL6nOR5UnIHSYEXX" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/ohw8OEJFmiH3YQ8zRnIw" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/9VnRKNuZ3sK26oyh0jX7" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/BhsY0CrwajITVEDsE4kP" alt="" width="563"><figcaption></figcaption></figure>

* **Trellis2** – New 3D generation backend

<figure><img src="/files/JA6aKhX5zitOTEuIIcwX" alt="" width="528"><figcaption></figcaption></figure>

<figure><img src="/files/LVYlIJ1SQQLaEs1MMZf7" alt="" width="563"><figcaption></figcaption></figure>

* **Hitem3D v2.0** – Upgraded backends with v1.5 retexture support

<figure><img src="/files/dAGQSiEu9M8OCUh4jacH" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/ziqQVKeaR8wj3hS4slH4" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/ZhHfyWPWkYJ3ERYZqiwq" alt="" width="563"><figcaption></figcaption></figure>

* **Other Multiview Backends** – Extended multiview generation options

<figure><img src="/files/xgVooUBPqa5dAfJOIZaZ" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/Me4tZFBHZvjs75rnnp7M" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/YCDF1RvfHkbeFXZax3FU" alt="" width="563"><figcaption></figcaption></figure>

#### Utility Nodes <a href="#utility-nodes" id="utility-nodes"></a>

* **Extract Image Layers Node** – Uses Qwen Image Layered for layer extraction

<figure><img src="/files/nenKjXurZeHqynpButX2" alt="" width="542"><figcaption></figcaption></figure>

* **Video Models Update** – Enhanced video generation capabilities

<figure><img src="/files/oHME7S2tLOSvaNBXfgkS" alt="" width="563"><figcaption></figcaption></figure>

* **Image Description** – Migrated to Gemini 3 Flash with minimal thinking

<figure><img src="/files/nEck2nVDGJb95JQkBmCw" alt="" width="563"><figcaption></figcaption></figure>

***

### 2. Conversational Agent / Assistant <a href="#id-2.-conversational-agent-assistant" id="id-2.-conversational-agent-assistant"></a>

#### Platform Agent (New) <a href="#platform-agent-new" id="platform-agent-new"></a>

* **Conversational Platform Agent** – Fully revamped AI assistant that can help users build pipelines through natural language conversation

<figure><img src="/files/k3KyG9MM2CIk9jhaDzh9" alt="" width="563"><figcaption></figcaption></figure>

* **Agent Context Awareness** – Improved understanding of platform usage and available features

<figure><img src="/files/JB4eVeXP0VEb8BlUk69U" alt="" width="563"><figcaption></figcaption></figure>

* **API Export Awareness** – Agent now understands and can guide users through API export functionality

{% columns %}
{% column %}

<figure><img src="/files/JxfGBN6CyoyeOGtyzphd" alt="" width="390"><figcaption></figcaption></figure>
{% endcolumn %}

{% column %}

<figure><img src="/files/JxfGBN6CyoyeOGtyzphd" alt="" width="390"><figcaption></figcaption></figure>
{% endcolumn %}
{% endcolumns %}

<figure><img src="/files/CvSGc4nYJfQHq9Kduamz" alt="" width="563"><figcaption></figcaption></figure>

* **Suggestion Engine Improvements** – Better AI-generated node suggestions with output tokens limit

<figure><img src="/files/c1A4VBj2by2JrgOf53vv" alt="" width="389"><figcaption></figcaption></figure>

* **Vastly Improved Node Placement** – Enhanced heuristics for AI-generated node positioning

#### Chat Interface <a href="#chat-interface" id="chat-interface"></a>

* **Auto-resizing Chatbot Textarea** – Dynamic input field that grows with content

<figure><img src="/files/fd0X1qUxYrgsB56jUnjI" alt="" width="394"><figcaption></figcaption></figure>

* **Revamped Sidebar** – Cleaner integration of the chat panel with the main interface

<figure><img src="/files/xY93EQSq3r9LNyIWiIZo" alt="" width="563"><figcaption></figcaption></figure>

***

### 3. UI Changes <a href="#id-3.-ui-changes" id="id-3.-ui-changes"></a>

#### Major Frontend Rewrite <a href="#major-frontend-rewrite" id="major-frontend-rewrite"></a>

* **Vue Flow Migration** – Complete graph system rewrite using Vue Flow for significant performance improvements
* **Graph Speedup & Optimization** – Faster node/edge rendering and manipulation
* **Tailwind CSS Integration** – Modernized styling with Tailwind, removed redundant classes

#### User Experience <a href="#user-experience" id="user-experience"></a>

* **Onboarding Video Carousel** – Interactive carousel with Stripe integration for new users

<figure><img src="/files/lUIqSioRdRg0GNbb2SLU" alt="" width="563"><figcaption></figcaption></figure>

* **User Credit Management View** – Dedicated interface for managing credits

<figure><img src="/files/TT46Uwvaq71osvS3Ug5t" alt="" width="563"><figcaption></figcaption></figure>

* **Promo Banner System** – Promotional banner with trigger logic
* **3D Viewer Loading State** – Visual feedback during mesh loading

<figure><img src="/files/QHz6M1sKNPn09tvE6MZd" alt="" width="563"><figcaption></figcaption></figure>

* **Mesh Preview Pending Spinner** – Loading indicator for mesh previews

<figure><img src="/files/HQwmsPVEQw2za9mq3MME" alt="" width="416"><figcaption></figcaption></figure>

* **Mobile Lock Page** – Improved mobile experience with lock page
* **Subscriptions Order & Labels** – Better organization of subscription tiers
* **Page Editor Colors** – Enhanced visual differentiation
* **Frontend Restyle** – General UI polish and consistency improvements
* **Contact Us Button** – Added to TosFooter component
* **Legacy Stop Button Removed** – Cleaned up deprecated UI elements

#### Performance <a href="#performance" id="performance"></a>

* **Browser Cache Headers** – Added caching for file endpoints
* **Removed Loading Hint** – Removed "This may take a few seconds to load" unnecessary message

***

### Migration Notes <a href="#migration-notes" id="migration-notes"></a>

* **Breaking Changes:** None expected for end users
* **API Changes:** The exported API now uses billing; ensure adequate credits
* **Performance:** Frontend graph operations should be noticeably faster due to Vue Flow migration

### New Experimental Nodes <a href="#new-experimental-nodes" id="new-experimental-nodes"></a>

#### (These nodes need to be enabled by request) <a href="#these-nodes-need-to-be-enabled-by-request" id="these-nodes-need-to-be-enabled-by-request"></a>

* **Mesh Multi-View Render Node** – Renders meshes from multiple viewing angles

<figure><img src="/files/UJ7OAcWivNLWJddJT0kR" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/6aDi6aKN2Ijtp9Egzdyx" alt="" width="563"><figcaption></figcaption></figure>

* **Multi-View Projection Node** – Projects multi-view images onto mesh surfaces

<figure><img src="/files/tckAdJq6EMsgU9OpJFEO" alt="" width="563"><figcaption></figcaption></figure>

* **Occlusion Baking Node** – Bakes ambient occlusion into mesh textures

<figure><img src="/files/zM4Bf8GwcylPMUKK1Vr7" alt="" width="563"><figcaption></figcaption></figure>

### Contributors <a href="#contributors" id="contributors"></a>

Thanks to all contributors who made this release possible across 137 commits spanning almost 2 months of development.


# Release 0.16.0

## NOTE <a href="#note" id="note"></a>

This release note includes all changes since 0.14.0.

**Release Date:** February 2026

**Previous Version:** v0.14.0 (January 2026)

***

### 1. New Views & Interface Changes <a href="#id-1.-new-views-and-interface-changes" id="id-1.-new-views-and-interface-changes"></a>

#### New Pages <a href="#new-pages" id="new-pages"></a>

* **Exported APIs** — Dedicated dashboard to manage exported APIs: view status, disable/enable, delete, and see code examples for integration. Accessible from the user menu.

<figure><img src="/files/SWCb1NBG5We3XfCka7TI" alt="" width="563"><figcaption></figcaption></figure>

* **Transactions** — Billing transaction history view showing credit usage per node execution. Role-gated (can be hidden or force-shown per user role).

<figure><img src="/files/mL7sXhmTTZ06GdmDlLRj" alt="" width="563"><figcaption></figcaption></figure>

* **Security Settings** — New "Security" tab under user profile for changing password.

<figure><img src="/files/5HkggUfMyVdMsIOLa5Ul" alt="" width="563"><figcaption></figcaption></figure>

#### Full Visual Redesign (3 iterations) <a href="#full-visual-redesign-3-iterations" id="full-visual-redesign-3-iterations"></a>

* Complete overhaul of the graph editor look and feel — new typography (TWKEverett Mono), redesigned nodes, edges, fields, slots, and dropdowns.
* Sidebar redesigned with new layout, project cards show single cover image with description and improved date formatting.

<figure><img src="/files/yawHIY78t6MEk6zP2FTS" alt="" width="563"><figcaption></figcaption></figure>

* New standardized UI component library (buttons, inputs, selects, panels, tabs, toggles, badges, etc.) applied consistently across all views.
* Deleted the old sidebar and toolbar — navigation consolidated into the user menu.

#### Authentication <a href="#authentication" id="authentication"></a>

* **Google Sign-In** — Users can now log in with their Google account in addition to email/password.

<figure><img src="/files/PSC6eFzghVqV77gASp3G" alt="" width="563"><figcaption></figcaption></figure>

#### Graph Editor Improvements <a href="#graph-editor-improvements" id="graph-editor-improvements"></a>

* **Project auto-fit on load** — Graph automatically fits to viewport when opening a project.
* **Sidebar is now resizable** by dragging.
*

```
<figure><img src="../../.gitbook/assets/Adsız tasarım (3) (1).gif" alt="" width="563"><figcaption></figcaption></figure>
```

* **Improved node search** — Better ranking and matching when searching for nodes in the creation menu.
* **Agent chat streaming** — The AI assistant now shows live tool calls, thinking indicators, and streaming text in real time instead of waiting for the full response.
* **Richer agent tool call display** — Tool calls show arguments, results, and status inline in the chat panel.

<figure><img src="/files/L3B77J1BELVvdXD357i2" alt="" width="563"><figcaption></figcaption></figure>

#### Fixes & Polish <a href="#fixes-and-polish" id="fixes-and-polish"></a>

* Fixed zoom scaling on multiline text and sidebar at different zoom levels.
* Audio/video players now stop when a node is re-run; scrub/seek errors suppressed.
* Terms of Service and Privacy Policy pages properly styled with article formatting.
* Fixed text undo in node text fields.
* Hotkeys no longer trigger while typing in input fields.
* Fixed node cost/price display.
* Fixed node sizing for older projects opened after the redesign.
* Re-added scrolling for image array fields.
* Fixed project name editing.
* Fixed onboarding dialog and socket issues.
* New project empty state redesigned.

***

### 2. Changes to Existing Nodes & New Backends/Models <a href="#id-2.-changes-to-existing-nodes-and-new-backends-models" id="id-2.-changes-to-existing-nodes-and-new-backends-models"></a>

#### New Backends <a href="#new-backends" id="new-backends"></a>

* **Seedream 3.0** added as a backend for advanced image generation (replacing the old Replicate-based path).
* **404.xyz (Four-o-Four)** added as a new backend for Image-to-3D, with clear error messages for unsupported options (PBR, face count).

<figure><img src="/files/54AQg0nv1Ox71hT3mbgZ" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/yhev9Aa6ahpeLJuutMRS" alt="" width="563"><figcaption></figcaption></figure>

* **ElevenLabs via** [Generative AI | Run Image, Video, 3D and Audio Models | fal.ai](http://fal.ai/) — ElevenLabs audio services now available through [Generative AI | Run Image, Video, 3D and Audio Models | fal.ai](http://fal.ai/) proxy for lower latency, with direct API fallback.

<figure><img src="/files/03g0bA9M97pBVDWT5TEw" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/heihKj6IwFtzYeIxuQlV" alt="" width="563"><figcaption></figcaption></figure>

* **Gemini native SDK** — Gemini models now use the native `google.genai` SDK with automatic fallbacks to [Generative AI | Run Image, Video, 3D and Audio Models | fal.ai](http://fal.ai/) and Replicate if the primary provider fails.

#### Existing Node Improvements <a href="#existing-node-improvements" id="existing-node-improvements"></a>

* **Image generation (advanced)** — Overhauled aspect ratio handling; added workarounds for Kling v2.1 and Reve API input size/resolution limitations (automatic compression and resizing).

<figure><img src="/files/LytPb63QcX2bqxUhJzrk" alt="" width="563"><figcaption></figcaption></figure>

* **LLM / Text node** — New **structured output mode** that returns validated JSON matching a user-defined schema. Pruned model list to remove underperforming models; small models blocked from complex tasks.

<figure><img src="/files/Z8yjjVnUcRV1hmZ8Ch8l" alt="" width="563"><figcaption></figcaption></figure>

* **3D scene composition** — Rendering distance now derived automatically from the scene instead of requiring manual input.
* **Remesh node** — Expanded with additional options.

<figure><img src="/files/hoj95QGI7ltKm3oWKBwc" alt="" width="563"><figcaption></figcaption></figure>

* **Node execution timeout** — Nodes that run too long are now automatically timed out instead of hanging indefinitely.
* **Node interruption** — Users can interrupt running nodes; stuck nodes are automatically detected and recovered.

<figure><img src="/files/X2DlVLh5iVU24MteTDbG" alt="" width="471"><figcaption></figcaption></figure>

***

### 3. New Nodes <a href="#id-3.-new-nodes" id="id-3.-new-nodes"></a>

#### Audio (10 nodes) <a href="#audio-10-nodes" id="audio-10-nodes"></a>

| Node                            | What it does                                             |
| ------------------------------- | -------------------------------------------------------- |
| **Audio SFX**                   | Generate sound effects from text descriptions            |
| **Audio Tags Enrichment**       | AI-powered enrichment of audio tags/metadata             |
| **Composition Plan**            | Plan an audio composition (structure, instruments, mood) |
| **Create Music**                | Generate music from text prompts                         |
| **Text to Dialogue**            | Generate multi-speaker dialogue audio                    |
| **Text to Speech**              | Basic text-to-speech conversion                          |
| **Text to Speech (ElevenLabs)** | High-quality TTS via ElevenLabs voices                   |
| **Voice Changer**               | Convert/clone voices                                     |
| **Voice Preview**               | Preview and select from available voices                 |

<figure><img src="/files/ZsTu1vuatOESrmmxdZkI" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/HFT4cuSf4ZtjKyXvWvyJ" alt="" width="563"><figcaption></figcaption></figure>

#### NovelAI — Experimental (16 nodes) <a href="#novelai-experimental-16-nodes" id="novelai-experimental-16-nodes"></a>

| Node                       | What it does                                              |
| -------------------------- | --------------------------------------------------------- |
| **Text to Image**          | Generate images using NovelAI's anime/illustration models |
| **Img2Img**                | Transform existing images with NovelAI                    |
| **Inpainting**             | Paint over parts of an image and regenerate               |
| **Vibe Transfer**          | Transfer the style/vibe from one image to another         |
| **Encode Vibe**            | Encode an image's vibe for use in vibe transfer           |
| **Precise Reference**      | Generate images using precise reference images            |
| **Upscale**                | Upscale images using NovelAI                              |
| **Director Tools**         | Control camera angle, lighting, and composition           |
| **Multi-Character Prompt** | Build prompts for scenes with multiple characters         |
| **Danbooru Prompt**        | Generate Danbooru-style tag prompts via Gemini            |
| **Prompt Enhancement**     | Enhance and refine prompts                                |
| **Tag Suggestions**        | Get tag suggestions for image generation                  |
| **Technical Params**       | Fine-tune technical generation parameters                 |
| **Text Generation**        | Generate text/stories                                     |
| **Text to Speech**         | NovelAI voice synthesis                                   |

<figure><img src="/files/1QVVfCIq6g8zDcpXzud5" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/6nur7C1eXp50SbTIpTAD" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/xaX5jlwjJrLPCUP2vTHe" alt="" width="563"><figcaption></figcaption></figure>

#### Animation & Mesh (4 nodes) <a href="#animation-and-mesh-4-nodes" id="animation-and-mesh-4-nodes"></a>

| Node                        | What it does                                                                 |
| --------------------------- | ---------------------------------------------------------------------------- |
| **Input Animation**         | Import animation files (FBX, BVH, etc.)                                      |
| **Normalize Humanoid Mesh** | Normalize a humanoid mesh to a standard skeleton for animation compatibility |
| **Retarget Animation**      | Transfer an animation from one rig to another                                |
| **Video to Animation**      | Extract motion/animation data from a video clip (via Meshcapade)             |

<figure><img src="/files/RIgXrcznx8ihwPnt883c" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/wW4eBoMjSg3L2n2PvVdc" alt="" width="563"><figcaption></figcaption></figure>

#### Image Processing (10 nodes) <a href="#image-processing-10-nodes" id="image-processing-10-nodes"></a>

| Node                      | What it does                                          |
| ------------------------- | ----------------------------------------------------- |
| **Split Alpha**           | Separate an image into RGB and alpha channel          |
| **Combine Alpha**         | Merge an RGB image with an alpha mask                 |
| **Split Channels**        | Split an image into individual R, G, B channels       |
| **Combine Channels**      | Combine separate R, G, B channels into one image      |
| **Threshold Binary Mask** | Create a black/white mask from a brightness threshold |
| **Crop (Manual)**         | Crop an image with specific coordinates               |
| **Crop (Auto)**           | Auto-trim transparent or uniform borders              |
| **Resize**                | Resize an image to target dimensions                  |
| **Concatenate Images**    | Join multiple images side-by-side or vertically       |
| **Render Text**           | Render text onto an image                             |

<figure><img src="/files/Kh1XNPWa62WyFkVgCSJk" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/CwkvM7OJIoExlBVQCx4f" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/f3LkPrVrHcVBTw6Zq9Mo" alt="" width="563"><figcaption></figcaption></figure>

#### Other (2 nodes) <a href="#other-2-nodes" id="other-2-nodes"></a>

| Node                         | What it does                                                                               |
| ---------------------------- | ------------------------------------------------------------------------------------------ |
| **Normal from Depth**        | Convert a depth map into a normal map                                                      |
| **Procedural Image Builder** | Generate images procedurally via AI-written code (e.g., patterns, gradients, compositions) |

<figure><img src="/files/QQ2fXRKGi8eoU5jfm6wX" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/GYqWYD6h658UiZpSBhHF" alt="" width="549"><figcaption></figcaption></figure>

***

### 4. Other User-Facing Changes <a href="#id-4.-other-user-facing-changes" id="id-4.-other-user-facing-changes"></a>

* **AI Assistant improvements** — The agent can now edit node field values directly, has better knowledge of the platform, and its suggestion engines are faster and more context-aware across all content types (images, meshes, prompts, collections).

<figure><img src="/files/wbff8hXw2Hpl7HYhRbP5" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/E60KFRSY1YE1VED2UikC" alt="" width="563"><figcaption></figcaption></figure>

* **Unified image descriptions** — All image description generation now uses a single, consistent pipeline with configurable models (visible as better auto-descriptions on uploaded images).

<figure><img src="/files/OHxSuk3IICQFrDPQv3kp" alt="" width="563"><figcaption></figcaption></figure>

* **3D API timeouts extended** — 3D generation tasks now wait up to 30 minutes before timing out, accommodating complex meshes.
* **CORS fix for file downloads** — Resolved an issue where downloading files could fail due to browser cache collisions.
* The **"Schedule (Run Predecessors)"** button — tool for running an entire pipeline or a sequence of nodes at once.

<figure><img src="/files/tt5v8Eg2bKVntETY8TIL" alt="" width="563"><figcaption></figcaption></figure>


# Release 0.17.0

## NOTE <a href="#note" id="note"></a>

This release note includes all changes since 0.16.2.

**Release Date:** March 2026

**Previous Version:** v0.16.2 (February 2026)

**Updates:** Frontend Modifications, New Backend Options, and Audio Format Switch to MP3

***

### 1. Interface Modifications (Frontend) <a href="#id-1.-interface-modifications-frontend" id="id-1.-interface-modifications-frontend"></a>

**3D viewer field labels updated** — The Img-to-3D node's texture quality dropdown labels now reflect 404-GEN support: `"Standard (Tripo: standard, Trellis: 2048, 404: 1024)"` and `"Detailed (Tripo: high-res, Trellis: 4096, 404: 2048)"`.

<figure><img src="/files/2xJ3NjSjQu0CBXsV0aR3" alt="" width="363"><figcaption></figcaption></figure>

***

### 2. Changes to Existing Nodes & New Backends/Models <a href="#id-2.-changes-to-existing-nodes-and-new-backends-models" id="id-2.-changes-to-existing-nodes-and-new-backends-models"></a>

**Text Generation (LLM) Node** — Added `gemini-3.1-flash-lite` and `gpt-5.3-chat`

<figure><img src="/files/AyklXHlhxx75Skbq2fRf" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/i2jHoUgjYG5IaduGJ0Bj" alt="" width="563"><figcaption></figcaption></figure>

**Gemini 3.1 Flash Image** (#994) — Added `gemini-3.1-flash-image-preview` as a new backend option across three image generation nodes:

* **Multimodal Node** — New `"Gemini 3.1 Flash Image"` backend.

<figure><img src="/files/pXKoozQFjSVJlselOJxY" alt="" width="563"><figcaption></figcaption></figure>

* **ImageTextEdit Node** — Added Gemini 3.1 Flash as a backend for image editing.

<figure><img src="/files/dZgTvhIpwFnYmbBAAJg5" alt="" width="563"><figcaption></figcaption></figure>

* **QualityTextToImage Node** — Added Gemini 3.1 Flash as a text-to-image backend alongside existing Gemini 3 Pro and 2.5 Flash.

<figure><img src="/files/9GgagC7GnNGiUydW0IJL" alt="" width="451"><figcaption></figcaption></figure>

**404-GEN upgrade** — Major upgrade to the 404.xyz 3D mesh backend:

<figure><img src="/files/LB8YqQau9qFJVK9Fwuii" alt="" width="563"><figcaption></figcaption></figure>

* Previously rejected any `face_limit` — now validates and passes it through.
* PBR remains unsupported (hard error).

<figure><img src="/files/HbgboAaOnVihCuXz1udG" alt="" width="563"><figcaption></figcaption></figure>

* Trellis 2 also updated: minimum decimation target lowered from 100k to 20k, added PBR hard error (was silently ignored), added texture-off override warning.

<figure><img src="/files/5loEaGDkuScU9YSmd9G2" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/rShxdP71M34fyaZO0iFr" alt="" width="563"><figcaption></figcaption></figure>

* **Tripo3D v3.1** added

<figure><img src="/files/594jhmbxiPwGNOELStJI" alt="" width="563"><figcaption></figcaption></figure>

* **Meshy** — additional generation arguments added

<figure><img src="/files/KtzxtZn4uBPxNbXY5y6b" alt="" width="501"><figcaption></figcaption></figure>

* **Retexture Mesh Node** — added support for multiple style images; Tripo backend added; additional Meshy generation arguments added

<figure><img src="/files/yEI9VckdxiSWiyW3PFNs" alt="" width="563"><figcaption></figcaption></figure>

* **Migration** (v0.9.49) — Existing projects with 404-GEN or Trellis 2 backends and `pbr=True` are auto-migrated to `pbr=False` to prevent execution errors.

**Audio format switch to MP3** (#987) — Comprehensive refactor of the internal audio format from WAV to MP3 320 kbps:

* **All audio nodes** now output `.mp3` / `audio/mpeg` instead of `.wav` / `audio/wav`.
* Uploaded audio is now converted to MP3 (320 kbps) instead of WAV.

***

### 3. New Nodes <a href="#id-3.-new-nodes" id="id-3.-new-nodes"></a>

No entirely new node types were added in this period.

***


# Release 0.18.0

This release note includes all changes since 0.17.0.

**Release Date:** March 2026

**Previous Version:** v0.17.0. (March 2026)

***

### Agent improvements <a href="#agent-improvements" id="agent-improvements"></a>

#### The platform assistant has ben significantly improved, and is capable of creating complex workflows with one prompt: <a href="#the-platform-assistant-has-ben-significantly-improved-and-is-capable-of-creating-complex-workflows-w" id="the-platform-assistant-has-ben-significantly-improved-and-is-capable-of-creating-complex-workflows-w"></a>

<figure><img src="/files/2CtBFiy71N4XQ6mZyJyD" alt="" width="563"><figcaption></figcaption></figure>

### Video Generation <a href="#video-generation" id="video-generation"></a>

#### New backends & models <a href="#new-backends-and-models" id="new-backends-and-models"></a>

* **Runway Gen 4.5** added to the Simple Text + Image → Video node (5/10 second durations, 720p, text-to-video and image-to-video with seed support).

<figure><img src="/files/jQMz63dDHzOL2YgR2Ksu" alt="" width="563"><figcaption></figcaption></figure>

* **Kling Video 3.0 Pro** added to the Simple Text + Image → Video node (3–15 seconds, image-to-video with optional end frame and audio toggle).

<figure><img src="/files/oS2zX69d975IDo2Xo50H" alt="" width="563"><figcaption></figcaption></figure>

* **Sora 2 / Sora 2 Pro** added to the Text + Image → Video node alongside Veo and Grok.

<figure><img src="/files/CWPKkYV9QWA8UoJfYeCk" alt="" width="563"><figcaption></figcaption></figure>

* **Grok Imagine** video generation and editing support expanded across video nodes.

<figure><img src="/files/1UMPnPPfea64bLck2lK0" alt="" width="563"><figcaption></figcaption></figure>

#### New video capabilities <a href="#new-video-capabilities" id="new-video-capabilities"></a>

* **Video Extend** node: extend existing videos using Veo, Grok, or Pixverse backends.

<figure><img src="/files/hPPyRQ2LchbUK973bnIu" alt="" width="563"><figcaption></figcaption></figure>

* **Video Edit** node: edit videos with Grok Imagine or Kling-based motion paths.

<figure><img src="/files/WD2WXbSDij1k6fE0Ufzu" alt="" width="563"><figcaption></figcaption></figure>

* **End Image** input added to Seedance and Kling backends for image-to-video, enabling start-to-end frame control.

<figure><img src="/files/0vnDw0Jvpdw6IL38QLt6" alt="" width="563"><figcaption></figcaption></figure>

* Seedance resolution options expanded to 480p / 720p / 1080p.

<figure><img src="/files/SGyQS3Rvh3xOphQAbMIb" alt="" width="440"><figcaption></figcaption></figure>

#### Dynamic pricing for video <a href="#dynamic-pricing-for-video" id="dynamic-pricing-for-video"></a>

* Video nodes (Simple Text + Image → Video, Text + Image → Video, Video Extend, Video Edit) now use **dynamic pricing** — cost is calculated per run based on actual backend, duration, and resolution instead of a flat credit price.

<figure><img src="/files/jSdBAgFaUuZZRcvrsBuP" alt="" width="497"><figcaption></figcaption></figure>

* The node header in the graph shows **"dynamic (N credits hold)"** so you know the maximum reserved before a run, and **"last run: X"** after completion.

***

### 3D Generation <a href="#id-3d-generation" id="id-3d-generation"></a>

#### New backends & models <a href="#new-backends-and-models.1" id="new-backends-and-models.1"></a>

* **Tripo P1** and **Tripo P1 Multiview** added as new Image → 3D backends with dedicated pricing tiers.

<figure><img src="/files/yNpatyIzxI8cYUy9HoAo" alt="" width="563"><figcaption></figcaption></figure>

#### Upgrades & fixes <a href="#upgrades-and-fixes" id="upgrades-and-fixes"></a>

* **404-GEN** now always uses 2048 texture resolution (Standard quality override removed).

<figure><img src="/files/whOaVXqjLeFIRmGq0zk3" alt="" width="499"><figcaption></figcaption></figure>

* **Hitem3D** pricing tiers refined (default, pro, portrait).

<figure><img src="/files/1hYYucPC6RxutRels3pS" alt="" width="421"><figcaption></figcaption></figure>

* **Image → 3D (With Fallback)** node switched to dynamic pricing (100 credits hold).

<figure><img src="/files/29yyeOF1ldeLZ4DWhPSr" alt="" width="448"><figcaption></figcaption></figure>

***

### Background Removal <a href="#background-removal" id="background-removal"></a>

* **Pixelcut** added as the new default Remove Background backend.

<figure><img src="/files/pm65v5bJJ71PD4HcMAX6" alt="" width="563"><figcaption></figcaption></figure>

***

### Graph Editor & Node UX <a href="#graph-editor-and-node-ux" id="graph-editor-and-node-ux"></a>

#### Seed & cache behavior <a href="#seed-and-cache-behavior" id="seed-and-cache-behavior"></a>

* Nodes with a **randomizing seed** (seed = -1 / "Randomize") are now always treated as cache-disabled — the Run button stays active and a fresh random seed is generated each execution. Applies to both `seed` and `texture_seed` fields.

<figure><img src="/files/PQ2UiLSLQAZEKhrD3KoS" alt="" width="563"><figcaption></figcaption></figure>

#### Creation menu <a href="#creation-menu" id="creation-menu"></a>

* Node groups where **all children are access-restricted** (e.g. behind a role gate) now appear visually dimmed in the creation menu, while remaining navigable so users can see what's available.

#### PDF viewer <a href="#pdf-viewer" id="pdf-viewer"></a>

* PDF preview in the file viewer now scales to **80% of viewport** (80vw × 80vh) with proper max constraints, replacing the previous awkward sizing.

<figure><img src="/files/uCLDTQjMVnQ7Yt8aJ1Mk" alt="" width="563"><figcaption></figcaption></figure>

#### Dynamic pricing display <a href="#dynamic-pricing-display" id="dynamic-pricing-display"></a>

* Nodes with dynamic pricing show the **credit hold amount** and **last run cost** directly on the node in the graph.
* Balance check uses the hold value for dynamic nodes to determine if a run is affordable.

***

### Help & Contact <a href="#help-and-contact" id="help-and-contact"></a>

* **Help** button in the user menu now links to the external documentation site (docs.atlas.design).

<figure><img src="/files/Y1HQNKBcn2pNCvdzOFmR" alt="" width="462"><figcaption></figcaption></figure>

<figure><img src="/files/nNHxhabpPGIIwmNbVooL" alt="" width="563"><figcaption></figcaption></figure>

* **Contact Us** option added to the user menu (opens email to <support@atlas.design>).

<figure><img src="/files/cxdjgAirie2N0axVZEwp" alt="" width="563"><figcaption></figcaption></figure>


# Release 0.19.0

**NOTE**\
This release note includes only Workspace-related changes introduced in v0.19.0.

**Release Date:** April 2026\
**Previous Version:** v0.18.0 (March 2026)

**Updates:** Shared Workspaces, Collaboration, and Billing Controls

***

### 1. Shared Workspaces <a href="#id-1.-shared-workspaces" id="id-1.-shared-workspaces"></a>

**Shared Workspaces** — Introduced a new workspace system that enables teams to collaborate within a shared environment.

<figure><img src="/files/iL24iiqVojmxeJRs2oQK" alt="" width="455"><figcaption></figcaption></figure>

* Users can create dedicated workspaces and invite members via email

<figure><img src="/files/NT3brM87AIyr3jFHhk0B" alt="" width="546"><figcaption></figcaption></figure>

<figure><img src="/files/fMoTiWpZDtyiYZCsKP6c" alt="" width="563"><figcaption></figcaption></figure>

* Invited users are automatically added upon accepting the invitation link
* All members can access and work on projects within the same workspace

**Real-Time Collaboration** — Multiple users can work simultaneously on projects within a shared workspace environment (but it’s still experimental)

***

### 2. Roles & Permissions <a href="#id-2.-roles-and-permissions" id="id-2.-roles-and-permissions"></a>

**Workspace Roles** — Workspace owners can assign and manage different permission levels:

* **Admin** — Full control over the workspace, including:
  * Managing members
  * Editing workspace name
  * Setting billing and usage limits
* **Member** — Standard access to workspace projects without billing visibility

<figure><img src="/files/cZzdRRYlN3B71VF92WBc" alt="" width="512"><figcaption></figcaption></figure>

***

### 3. Billing & Usage Management <a href="#id-3.-billing-and-usage-management" id="id-3.-billing-and-usage-management"></a>

**Workspace Credit Limits** — Introduced flexible credit control mechanisms to manage usage:

* Limits define how many credits a workspace can spend within a billing period
* When the limit is reached, execution is blocked until the next usage period

**Usage Period Definition** —

* Usage is tracked in **rolling 30-day windows (UTC)**
* At the end of each period:
  * Per-member credit usage resets
  * Limits are reapplied for the new period

***

### 4. Limit Configuration Options <a href="#id-4.-limit-configuration-options" id="id-4.-limit-configuration-options"></a>

**Workspace-Wide Limits** —

* Apply to all members by default
* Define a global credit cap for the entire workspace

**Per-Member Limits** —

* Can override workspace-wide limits for specific users
* Allows granular control of individual usage

<figure><img src="/files/vHzfpElKkFP9qRNO2Vhc" alt="" width="563"><figcaption></figcaption></figure>

***

### 5. Shared Resources <a href="#id-5.-shared-resources" id="id-5.-shared-resources"></a>

**Shared API Access** — Exported workflow APIs are now visible and accessible within the shared workspace, enabling team-wide integration and reuse.

<figure><img src="/files/bfcABoVXQDg9CiKXvGHG" alt="" width="563"><figcaption></figcaption></figure>

***

### 6. Notes <a href="#id-6.-notes" id="id-6.-notes"></a>

This release introduces the foundation for **team-based collaboration in Atlas Studio** through Shared Workspaces, including role management, billing controls, and shared resource access. Other feature updates are not included in this note.


# Release 0.20.0

**Release Date:** April 2026

**Previous Version:** v0.19.0. (April 2026)

### New Features & Nodes <a href="#new-features-and-nodes" id="new-features-and-nodes"></a>

#### Seedance 2.0 Migration + Algorithmic Image Filters Node <a href="#seedance-2.0-migration--algorithmic-image-filters-node" id="seedance-2.0-migration--algorithmic-image-filters-node"></a>

* **Seedance 2.0** new **Reference to Video** node accepting up to 9 images, 3 videos, and 3 audio clips.

<figure><img src="/files/hwfv2bLbg9LixRctm1Wf" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/NAEV8JlB0Kuem9UkMTtJ" alt="" width="563"><figcaption></figcaption></figure>

* **69 algorithmic image filters** in a new Image Filters node across 8 categories (Noise, Blur, Color/Tone, Sharpen, Distortion, Degrade/Stylize, Weather, Blend Modes). Users get a single dropdown with built-in parameter guidance.

<figure><img src="/files/FosHOztvogdzFaLOd1ob" alt="" width="437"><figcaption></figcaption></figure>

<figure><img src="/files/9oXhtBYialf0ogLREWXG" alt="" width="563"><figcaption></figcaption></figure>

#### Node Field Visibility Conditions <a href="#node-field-visibility-conditions" id="node-field-visibility-conditions"></a>

* Fields on nodes now dynamically show/hide based on other field values (e.g., backend-specific parameters only appear when that backend is selected). "Backend" labels renamed to "Engine" across the UI.

<figure><img src="/files/M7DFEvMtIjSTMszqEVmn" alt="" width="563"><figcaption></figcaption></figure>

***

### Improvements & Quality of Life <a href="#improvements-and-quality-of-life" id="improvements-and-quality-of-life"></a>

#### Normals Mode on High-to-Low Poly Baking <a href="#normals-mode-on-high-to-low-poly-baking" id="normals-mode-on-high-to-low-poly-baking"></a>

* New `normals_mode` parameter (preserve / smooth / flat) on the bake node, giving users explicit control over low-poly shading before baking.

<figure><img src="/files/LCgYVYmgi8VAXrWg5xYn" alt="" width="563"><figcaption></figcaption></figure>

#### Insufficient Credits Display Fix <a href="#insufficient-credits-display-fix" id="insufficient-credits-display-fix"></a>

* Removed a faulty condition that could hide the "insufficient credits" message in the agent chat panel. Also cleaned up dead `axios` dependency.

#### Image Format Reliability Fix <a href="#image-format-reliability-fix" id="image-format-reliability-fix"></a>

* Third-party backends occasionally returned JPEG instead of PNG. All image outputs are now normalized to real PNG before metadata embedding. Extract PNG Metadata now shows the actual error instead of a generic "no metadata" message.

***

### Summary for in-browser users <a href="#summary-for-in-browser-users" id="summary-for-in-browser-users"></a>

The headline changes a user would **see and feel**:

1. **Seedance 2.0 video generation** — new Reference-to-Video node
2. **69 image filters** — blur, grain, color grading, blend modes, etc. — directly in the node graph
3. **Smart field visibility** — node panels are cleaner; irrelevant fields hide automatically based on selected engine
4. **\~16x faster aspect ratio suggestions** — feels nearly instant now
5. **Normals mode** on baking — new creative control
6. **Bug fixes** — correct suggestion values, reliable PNG output, proper credit warnings


# Release 0.21.0

**Release Date:** April 2026

**Previous Version:** v0.20.0. (April 2026)

### Billing & Workspaces <a href="#billing-and-workspaces" id="billing-and-workspaces"></a>

#### Team workspace billing and creation flow <a href="#team-workspace-billing-and-creation-flow" id="team-workspace-billing-and-creation-flow"></a>

* New **workspace creation page** (`workspace/add`) with tier selection at creation time.

<figure><img src="/files/chMsILMIG91bPoxNfMsE" alt="" width="563"><figcaption></figcaption></figure>

* Paid tiers go through a **“pending payment” state** until Stripe confirms the subscription, at which point the workspace becomes active.
* **Usage limits page** and **subscription management** added to workspace/profile settings; subscription routes consolidated under `billing:*`.

<figure><img src="/files/QassKNHFGpGfNkmbeXmv" alt="" width="563"><figcaption></figcaption></figure>

* Workspace-related UI is improved

#### Workspace/subscription cancellation <a href="#workspace-subscription-cancellation" id="workspace-subscription-cancellation"></a>

* Eligible users can **cancel a subscription** from the billing UI via a confirmation dialog.
* Cancelled-but-active subscriptions now show a **“cancelling on …”** state until the period ends.

***

### Projects View <a href="#projects-view" id="projects-view"></a>

#### Filter bar on projects view <a href="#filter-bar-on-projects-view" id="filter-bar-on-projects-view"></a>

* New filter strip above the project list: **User** (workspace mode), **Name substring**, **Date From/To**.

<figure><img src="/files/em3DLthUwg1k366MGUmw" alt="" width="563"><figcaption></figcaption></figure>

* Default selection is **“Me”** so users land on their own projects immediately; “All members” available as a second option.

<figure><img src="/files/f59tWDdRTNuxVaHQT1Xj" alt="" width="563"><figcaption></figcaption></figure>

* New **“no matches”** empty state with a **Reset filters** button (distinct from the existing “No projects yet” empty state).

<figure><img src="/files/DvyIBZaUwfNZzcRhS6FO" alt="" width="563"><figcaption></figcaption></figure>

* New reusable `CalendarSelect` dropdown date picker — also replaces the native date inputs in the **Transactions** view for consistency.

<figure><img src="/files/RBzzCFRBvsryzLcH2CWW" alt="" width="563"><figcaption></figcaption></figure>

***

### Graph Nodes — New <a href="#graph-nodes-new" id="graph-nodes-new"></a>

#### Lipsync node <a href="#lipsync-node" id="lipsync-node"></a>

New v2 node with four selectable backends and per-backend UI fields:

| Backend            | Mode                                  | Notes                                                                                |
| ------------------ | ------------------------------------- | ------------------------------------------------------------------------------------ |
| `pixverse`         | Audio **or** built-in TTS (15 voices) | Dual-provider with automatic failover                                                |
| `sync3`            | Audio-only                            | 5 sync modes for duration mismatch (`cut_off`, `loop`, `bounce`, `silence`, `remap`) |
| `heygen_speed`     | Audio-only                            | Fast dubbing                                                                         |
| `heygen_precision` | Audio-only                            | High-fidelity avatar dubbing                                                         |

<figure><img src="/files/bKIsButghCqYq3lkHJ8q" alt="" width="413"><figcaption></figcaption></figure>

<figure><img src="/files/Xaf4JpkJpS8whdnPw94H" alt="" width="563"><figcaption></figcaption></figure>

#### ERNIE image, ReconViaGen, PATINA Material Generation (#1172) <a href="#ernie-image-reconviagen-patina-material-generation-1172" id="ernie-image-reconviagen-patina-material-generation-1172"></a>

* **ERNIE image** added as a backend to:
  * `fast_text_to_image` → new `ernie-turbo` backend.
  * `quality_text_to_image` → new `ernie` backend.
  * Exposes negative prompt, inference steps, guidance scale, and prompt expansion. Warns when a parameter is ignored instead of silently dropping it.

<figure><img src="/files/HxslqNM93Xbi7ZTYjG72" alt="" width="563"><figcaption></figcaption></figure>

* **ReconViaGen 0.5** multiview → 3D added as a backend to `image_to_3d` and `multiview_to_3d` (new `reconviagen` / `reconviagen_mv` backends), with a new **multi-image strategy** control and PBR-always output.

<figure><img src="/files/VUl68oILrxNpkn1MfK2g" alt="" width="279"><figcaption></figcaption></figure>

* **Material Generation** — brand-new `material_generation` node backed by PATINA, with **generate / predict / extract** modes and per-map outputs.

<figure><img src="/files/zXRvpA6n8YL7AVvek5Mm" alt="" width="462"><figcaption></figcaption></figure>

<figure><img src="/files/FAR8oIt4R9MDCLWUMojj" alt="" width="563"><figcaption></figcaption></figure>

#### Remesh: in-cluster GPU backend (for now on request, will be globally enabled soon) <a href="#remesh-in-cluster-gpu-backend-for-now-on-request-will-be-globally-enabled-soon" id="remesh-in-cluster-gpu-backend-for-now-on-request-will-be-globally-enabled-soon"></a>

* `RemeshNode` gains a new `atlas_remesh` backend option, running on the in-cluster GPU remesher.
* `target_polycount` and `topology` fields are hidden for this backend (voxel remesher ignores them).

<figure><img src="/files/8BJOaBKrCMwviN59duqa" alt="" width="563"><figcaption></figcaption></figure>

***

### Graph Nodes — Updates <a href="#graph-nodes-updates" id="graph-nodes-updates"></a>

#### Claude Opus 4.6 → 4.7 <a href="#claude-opus-4.6-4.7" id="claude-opus-4.6-4.7"></a>

* `anthropic claude-opus-4-6` replaced with `anthropic claude-opus-4-7` across all nodes exposing a model selector: `simple_llm`, `structured_output`, `image_to_text`, `image_array_to_text`, `scene_composition_3d`, `procedural_image_builder`.

<figure><img src="/files/Ky21X1KPrB9I5iMRDkrO" alt="" width="455"><figcaption></figcaption></figure>

#### Bake High-Poly to Low-Poly: expose texture resolution <a href="#bake-high-poly-to-low-poly-expose-texture-resolution" id="bake-high-poly-to-low-poly-expose-texture-resolution"></a>

* New `texture_resolution` input on `bake_highpoly_to_lowpoly` — non-normal PBR map resolution is no longer silently auto-detected from the high-poly.
* Bake margin now scales with resolution (previously a fixed 16 px), giving cleaner seams at higher resolutions.

<figure><img src="/files/gtR5fWow3mDmRHQPbkS3" alt="" width="392"><figcaption></figcaption></figure>


# Release 0.22.0

**Release Date:** April 2026

**Previous Version:** v0.21.0. (April 2026)

**Updates:** Canvas annotations, cleaner node controls, new image and audio generation options, and self-serve subscription cancellation.

***

#### Organize Projects With Canvas Annotations <a href="#organize-projects-with-canvas-annotations" id="organize-projects-with-canvas-annotations"></a>

You can now add visual annotations directly to the project graph:

* Add groups around related nodes with the shortcut Ctrl+G
* Add sticky notes for comments, instructions, or reminders with the shortcut Ctrl+Alt+N.
* Add larger sections to structure complex canvases with the shortcut Ctrl+Alt+S.
* Edit annotation text, color, size, and font size from the canvas context menu.
* Move, resize, select, copy, cut, paste, delete, undo, and redo annotations alongside normal graph work.

<figure><img src="/files/gsOFEZlMnhxp49UNV6J7" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/6oHNBNYmKqRjmItE6JJB" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/dWg0cc8VcrLqKJNcCpX4" alt="" width="563"><figcaption></figcaption></figure>

#### Cleaner Advanced Controls on Nodes <a href="#cleaner-advanced-controls-on-nodes" id="cleaner-advanced-controls-on-nodes"></a>

Nodes now use a shared collapsible `Advanced` section for complex controls instead of separate per-node "show advanced" toggles. This keeps the default node view simpler while still preserving access to detailed settings when needed.

<figure><img src="/files/tRpoDV9kSR0JEJcC2Q1w" alt="" width="563"><figcaption></figcaption></figure>

#### Image Generation Updates <a href="#image-generation-updates" id="image-generation-updates"></a>

* OpenAI image generation has moved to GPT-Image-2, with selectable 1K, 2K, and 4K variants where available.

<figure><img src="/files/7jFS3THr8dm9Suc6GG2u" alt="" width="512"><figcaption></figcaption></figure>

* GPT-Image-2 is available in text-to-image and text-plus-image image generation flows.

<figure><img src="/files/FiAc0wsK7ki6Yh4eynvR" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/7ntXx0GdjbdjWz3FNU3k" alt="" width="563"><figcaption></figcaption></figure>

* Text-plus-image generation now supports GPT-Image-2 with up to 10 reference images.

<figure><img src="/files/hdKCvMt8pg6NDlafIzlu" alt="" width="563"><figcaption></figcaption></figure>

* Qwen Image 2 Pro is now available for text-to-image, image editing, multimodal image composition, and image extension.

<figure><img src="/files/Xdt4FaiTfumfl8mb8f2w" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/lRo5cDr5MO5dbvX4n0xU" alt="" width="563"><figcaption></figcaption></figure>

* The Extend Image node now offers multiple engines, including FLUX Fill, Gemini 3.1 Flash Image, and Qwen Image 2 Pro.

<figure><img src="/files/ypRTa9GYH4og2kIbGvaC" alt="" width="563"><figcaption></figcaption></figure>

* For workspaces with restricted image-model access, unavailable engines are hidden or routed to an allowed alternative with a visible warning when applicable.

#### New Audio Generation and Editing Options <a href="#new-audio-generation-and-editing-options" id="new-audio-generation-and-editing-options"></a>

* New Simple Music Generation node for generating music with MiniMax Music v2.6, ACE-Step, Google Lyria 3 Clip, and Google Lyria 3 Pro.

<figure><img src="/files/Wgs8upTratDGHyw6mMBW" alt="" width="525"><figcaption></figcaption></figure>

* New Simple Modify Music node for music covers, inpainting a time segment, remixing, and restyling existing audio.

<figure><img src="/files/F8U8r9UPKNBvRCvkqSQP" alt="" width="563"><figcaption></figcaption></figure>

* Simple Text to Speech now includes MiniMax Speech 2.8 HD and Gemini 3.1 Flash TTS in addition to the existing IndexTTS and xAI Grok TTS options.

<figure><img src="/files/O1mm7zzYDqqJSEfMLhPy" alt="" width="563"><figcaption></figcaption></figure>

* Audio tag enrichment can now target more downstream engines, including ElevenLabs, xAI, Gemini, MiniMax, and Lyria.

<figure><img src="/files/zzK0PM1oZUcKvsmJ5qUp" alt="" width="429"><figcaption></figcaption></figure>

#### Subscription Management <a href="#subscription-management" id="subscription-management"></a>

* Eligible users can now cancel an active subscription from the subscription details card.
* Cancellation uses a confirmation dialog and keeps access active until the end of the current billing period.
* The billing UI shows when a subscription is scheduled to cancel.
* Subscription plan, checkout, and top-up actions now respect role-based access in the UI.

### Summary <a href="#summary" id="summary"></a>

* Added graph annotations: groups, sticky notes, and sections for organizing canvases.
* Added a shared Advanced section for cleaner node controls.
* Upgraded OpenAI image generation to GPT-Image-2 variants.
* Added Qwen Image 2 Pro across image generation, editing, multimodal composition, and image extension.
* Added new music generation, music modification, and TTS engines.
* Added self-serve subscription cancellation for eligible users.


# Release 0.23.0

**Release Date:** May 2026

**Previous Version:** v0.22.0. (April 2026)

### New And Improved <a href="#new-and-improved" id="new-and-improved"></a>

#### Video <a href="#video" id="video"></a>

* Added more engines to the main text/image-to-video flow:
  * Veo 3.1 Lite for cheaper/faster Veo generation.
  * Vidu Q3 with t2v/i2v/r2v modes, audio, and movement controls.
  * PixVerse C1 with reference support, styles, and audio controls.
  * Wan 2.7 R2V with 720p/1080p, multi-shot support, reference media, and audio.
  * Kling O3 R2V Pro/4K with start/end frame support and Kling element references.
  * Kling Video 3.0 Pro/4K with first-frame generation, end frames, element references, multi-prompt segments, CFG scale, and audio controls.

<figure><img src="/files/rWb0ikuCFOadBlivtzkI" alt="" width="563"><figcaption></figcaption></figure>

* Improved Veo prompting support with clearer guidance for camera, subject, action, setting, style, and audio structure.

<figure><img src="/files/pNMV6GzRsTQprswWiLWb" alt="" width="563"><figcaption></figcaption></figure>

* Added or expanded video settings for durations, aspect ratios, resolutions, end frames, audio, negative prompts, multi-shot, auto-fix, and backend-specific visibility.

<figure><img src="/files/1yk0HGCEt6Izxl9tUe4s" alt="" width="563"><figcaption></figcaption></figure>

* Added Wan 2.7 to the Simple Text + Image to Video node, including t2v/i2v, native audio, end frame support, negative prompts, and prompt expansion.
* Improved Simple Video support for Kling 4K output and clearer resolution handling.

<figure><img src="/files/AuHum8PS1uIjZufQs0e0" alt="" width="563"><figcaption></figcaption></figure>

* Expanded Video Edit with:
  * Kling O3 V2V Reference for style transfer from reference video/images.

<figure><img src="/files/UGbcqWdXwYGWBPk4KurE" alt="" width="563"><figcaption></figcaption></figure>

* Wan 2.7 Video Edit for instruction-based edits with optional style reference image.

<figure><img src="/files/zz6aOkXpV3Khc106oSKi" alt="" width="563"><figcaption></figcaption></figure>

* Better Kling element reference fields and clearer duration/resolution validation.
* Video Extend presents clearer engine-specific controls for Veo 3.1, Grok Imagine, and Pixverse v6 continuation workflows.

<figure><img src="/files/buOq2EAMDSoL3Ac7UScq" alt="" width="563"><figcaption></figcaption></figure>

* Seedance 2.0 Standard now supports 1080p, and reference validation is clearer across image, video, and audio references.

<figure><img src="/files/M495WT0hM193Tg9Atxjg" alt="" width="342"><figcaption></figcaption></figure>

#### SVG <a href="#svg" id="svg"></a>

* Added a new Text to SVG node for prompt-based vector asset generation.

<figure><img src="/files/gxMhKfd35Z8P3CsyumeN" alt="" width="563"><figcaption></figcaption></figure>

* Renamed the previous image vectorization flow to Text + Image to SVG to better reflect prompted image/reference SVG generation.

<figure><img src="/files/j6URe2z3kj0NB9lxYhCw" alt="" width="563"><figcaption></figcaption></figure>

* Added Quiver Arrow 1.0, Arrow 1.1, and Arrow 1.1 Max as SVG backends.

<figure><img src="/files/h5FhwzLDNafEUcbGdr5R" alt="" width="563"><figcaption></figcaption></figure>

* Added Quiver-specific instructions, auto-crop, and target-size controls for SVG generation/vectorization.

<figure><img src="/files/MZbfTO2GvkHYRC2pxjlU" alt="" width="563"><figcaption></figcaption></figure>

#### Normal Map Convetion Convert Node <a href="#normal-map-convetion-convert-node" id="normal-map-convetion-convert-node"></a>

* Added a Normal Map Convention Convert node to flip normal maps between OpenGL (Y+ Up) and DirectX (Y- Down).

<figure><img src="/files/WAU8i57XYAadaCjuPTTc" alt="" width="563"><figcaption></figcaption></figure>

#### 3D, Meshes, And Materials <a href="#id-3d-meshes-and-materials" id="id-3d-meshes-and-materials"></a>

* Mesh texture tools now support ambient occlusion maps:
  * Extract Mesh Textures can output ambient occlusion.
  * Apply Mesh Textures can pack/apply ambient occlusion with roughness and metallic maps.

<figure><img src="/files/vpDVnonrtn179c1U3jE5" alt="" width="563"><figcaption></figcaption></figure>

* Apply Mesh Textures now accepts a normal map convention setting and converts DirectX maps to the glTF/OpenGL convention when needed.

<figure><img src="/files/UXfoRd6TY28qyZIH8LjB" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/Fbh9XUsJQ8Mb46JcCTtb" alt="" width="563"><figcaption></figcaption></figure>

* Bake Highpoly to Lowpoly now exposes AO Samples, giving users control over AO quality versus bake time.

<figure><img src="/files/kRXlbasaDDpt06VOJDt8" alt="" width="563"><figcaption></figcaption></figure>

* Bake Highpoly to Lowpoly now exposes Texture Margin, including an Auto option.

<figure><img src="/files/z1smZjJxO21ofgUmDmra" alt="" width="563"><figcaption></figcaption></figure>


# Release 0.24.0

**Release Date:** May 2026

**Previous Version:** v0.23.0. (May 2026)

### Creation Tools <a href="#creation-tools" id="creation-tools"></a>

* **Smart Resize Image** is now available in the Images node group for AI-assisted resizing and recomposition.

<figure><img src="/files/2sqT5e5IGnXcvOVsslYA" alt="" width="563"><figcaption></figcaption></figure>

* **Extend Image** now offers OneReward in addition to FLUX Fill, Gemini 3.1 Flash Image, and Qwen Image 2 Pro.

<figure><img src="/files/W8aS8tyBzTBauFdw2gf5" alt="" width="563"><figcaption></figcaption></figure>

* Quiver SVG generation now uses the newer Arrow 1.1 options; Arrow 1.0 has been removed from the SVG backend choices.

<figure><img src="/files/QDCPYG7F8KA0b29NltvX" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/irHFfmpdBhUqHiUSWkkk" alt="" width="563"><figcaption></figcaption></figure>

* HeyGen lipsync failures now show clearer guidance when no active speaker is detected, including cases where the speaker is visible but not speaking during the audio-covered timeframe.

<figure><img src="/files/k2w6yxS864DxDUI1lvQM" alt="" width="322"><figcaption></figcaption></figure>

### Projects And Collaboration <a href="#projects-and-collaboration" id="projects-and-collaboration"></a>

* Workspace project cards now show an **Access** selector for creators, making it easier to choose whether a project stays private, can be viewed by the workspace, or can be edited by the workspace.

<figure><img src="/files/HOZOti5rd6wjDDs3KreU" alt="" width="563"><figcaption></figcaption></figure>

* Workspace project access is enforced when opening, saving, and collaborating on projects, so read-only and editable workspace modes behave consistently.
* Shared and read-only projects now show a clearer **Save and Edit** dialog, guiding users to copy the project before making changes.

<figure><img src="/files/3ieBjIhGFOQM7qwitQDI" alt="" width="563"><figcaption></figcaption></figure>

* The project sidebar now keeps sharing controls available for editable projects and keeps copy-link actions available when a project is shared.

### Exported APIs <a href="#exported-apis" id="exported-apis"></a>

* Exported API project groups now distinguish projects by their source owner or workspace, avoiding collisions when different workspaces or users have projects with the same ID.

<figure><img src="/files/8a0ZIOIf7HCxSbco4L8I" alt="" width="549"><figcaption></figcaption></figure>


# Release 0.25.0

**Release Date:** May 2026

**Previous Version:** v0.24.0. (May 2026)

### New Creative Backends <a href="#new-creative-backends" id="new-creative-backends"></a>

* Added **Happy Horse** video support:
  * Text/image to video: 3-15s, 720p/1080p, native audio.
  * Reference-to-video: 1-9 reference images, 3-15s, 720p/1080p.
  * Video edit: 3-60s input videos, up to 5 image references, 720p/1080p, audio auto/original.

<figure><img src="/files/VVpSVSuv22BHLfaOM160" alt="" width="563"><figcaption></figcaption></figure>

* Added **Grok Imagine Quality 1K/2K** for text-to-image, image editing, and text+image-to-image workflows.

<figure><img src="/files/U8irfA74h84GBePxNa28" alt="" width="563"><figcaption></figcaption></figure>

* Added **Ideogram** as a background removal backend, with clearer fallback warnings when a selected backend changes provider.

<figure><img src="/files/kbqgZrsRAHl20RN87S0g" alt="" width="563"><figcaption></figcaption></figure>

### Agent Improvements <a href="#agent-improvements" id="agent-improvements"></a>

* The project agent can now run or schedule nodes, watch node completion, and report richer run outcomes.

<figure><img src="/files/Gc8YsiHY1b8P9mI7V8Ds" alt="" width="392"><figcaption></figcaption></figure>

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FCjfflFItcWlaWCoiZThl%2Fvideo_1281.mp4?alt=media&token=f8c882b9-83e0-474a-991a-86aa43f0bc71>" %}

* Added a per-project **Allow running nodes** toggle in agent chat.

<figure><img src="/files/2XoJflWywRwzmGkQ9FJk" alt="" width="383"><figcaption></figcaption></figure>

* Agent chat now has clearer timestamps, cleaner tool progress messages, better busy-state handling, and hides internal system messages for regular users.

<figure><img src="/files/PBppojJ3OWuLZ6QO3YkV" alt="" width="392"><figcaption></figcaption></figure>

<br>


# Release 0.28.0

> \
> **Release Date:** June 2026
>
> **Notes:** Cumulative, end-user-visible changes since **0.25.0** (covers 0.26.x and 0.27.x; no separate notes were published for the in-between releases).

***

### New & upgraded AI models <a href="#new-and-upgraded-ai-models" id="new-and-upgraded-ai-models"></a>

* **Grok Imagine Video 1.5** — new text + image-to-video engine (image-to-video, 1–15s, 480p/720p).

<figure><img src="/files/8zrC6xKuvcuZU3lZwiSr" alt="" width="563"><figcaption></figcaption></figure>

* **ByteDance Video Upscaler** — a new **Upscale Video** node that upscales clips up to 1080p/30fps.

<figure><img src="/files/reCvq0mQIIULwZKU37TK" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/p0BWDUOhO55BN7i3l1v8" alt="" width="563"><figcaption></figcaption></figure>

* **Hyper3D Rodin 2.5** — a direct image-to-3D and multi-view-to-3D backend with tier, texture-mode, Quad/Raw, face-limit, alpha, and de-light controls.

<figure><img src="/files/LxW9TiY4tWVSYpbBafhO" alt="" width="563"><figcaption></figcaption></figure>

* **Hitem3D Relief / Depth Pro** — generate depth/relief maps with native PNG previews and downloadable EXR output.

<figure><img src="/files/LA67nZOVOM7XMvn1anhI" alt="" width="563"><figcaption></figcaption></figure>

* **ControlFoley** — a new video-edit engine that generates synchronized foley/sound effects from an input video (with optional reference audio and prompts).

<figure><img src="/files/ndHIm0b8O9jx8T5O5VXW" alt="" width="563"><figcaption></figcaption></figure>

* **ElevenLabs upgrades** — selectable text-to-speech models with model-specific limits, plus a new **Speech to Text** node (Scribe v2).

<figure><img src="/files/CcUjolWOVtmxJoPMCWnw" alt="" width="563"><figcaption></figcaption></figure>

* **Recraft 4.1 Pro (Text to Vector)** — new selectable Text-to-SVG backend.

<figure><img src="/files/HXGSD9rQt1wFAkRzjJd5" alt="" width="563"><figcaption></figcaption></figure>

* **Google Gemini 3.5 Flash** — added as a first-class text/vision model across the editor and agent.

<figure><img src="/files/MASaEMXyEKjabqxhsT74" alt="" width="563"><figcaption></figcaption></figure>

* **OpenAI GPT-5.5 & GPT-5.5 Instant** — new model options; existing saved selections are migrated forward automatically.

<figure><img src="/files/xfiuJkjSijNvs8q7IIdZ" alt="" width="519"><figcaption></figcaption></figure>

* **Anthropic Claude Opus 4.8** — agent/LLM model upgraded from Opus 4.7 (saved graphs migrate automatically).

<figure><img src="/files/9jtRhjCTBB0DRGQBYNPH" alt="" width="563"><figcaption></figcaption></figure>

* **More reliable music generation** — Lyria 3 (Clip/Pro) gains an additional provider fallback for better uptime.

<figure><img src="/files/d3VAL0J2vkFL0aiTgWw6" alt="" width="563"><figcaption></figcaption></figure>

### New creative tools & nodes <a href="#new-creative-tools-and-nodes" id="new-creative-tools-and-nodes"></a>

* **Maya-compatible USDZ exchange** — new **Mesh → Maya USDZ** and **USDZ → Mesh** nodes with full USDZ upload/download support and material (base color, ORM, normal, UV) preservation.

<figure><img src="/files/lfpb4XUOgKIbjIw1uazp" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/LbiYBN62bW7GQO13q0L3" alt="" width="563"><figcaption></figcaption></figure>

* **Paste Crop Into Image** — paste an edited crop back into its source image, with alpha-aware compositing, resizing, and out-of-bounds clipping.

<figure><img src="/files/e0vCDKhxCxfcHlJKyE6x" alt="" width="563"><figcaption></figcaption></figure>

* **EXR support** — EXR files can be uploaded, previewed, and downloaded across the graph and API, including a new **EXR → 16-bit PNG** conversion node.

<figure><img src="/files/DYuZjK1xJ5N396HdnFpZ" alt="" width="457"><figcaption></figcaption></figure>

<figure><img src="/files/2BUZMDzIuvgv0B2mctyv" alt="" width="563"><figcaption></figcaption></figure>

### 3D preview & viewer <a href="#id-3d-preview-and-viewer" id="id-3d-preview-and-viewer"></a>

* **Studio-quality 3D preview** — the in-app 3D viewer now uses studio-style environment lighting, tone mapping, and antialiasing for far more accurate material color.

<figure><img src="/files/DWd9FdMYfgZQ5YS5QsVZ" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/ldgpKxQIBoO5xnkveBoC" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/HeVbDDH4h9ezki63XGsx" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/v9SGw4HFF2OE2jhaosUk" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/V2xfgXbd1DOhdqObJ5lP" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/TLJLe4MAGv6v3r4xpaFl" alt="" width="563"><figcaption></figcaption></figure>

* **Mesh animation controls** — pause/resume playback for animated 3D meshes directly in the viewer (Space bar toggles).

<figure><img src="/files/zXFPojIqYGffDE6wD7T6" alt="" width="563"><figcaption></figcaption></figure>

### Workspaces & collaboration <a href="#workspaces-and-collaboration" id="workspaces-and-collaboration"></a>

* **Easier team invites** — non-admin workspace members can now invite others (role-gated).

### Experience & UI <a href="#experience-and-ui" id="experience-and-ui"></a>

* **Redesigned dashboard & project view** — refreshed shell, card-based project grid, new toolbar (search/sort/filter), and modernized dialogs for a cleaner, faster home experience.

<figure><img src="/files/IxntMWDbvZ1st2RoJg9D" alt="" width="563"><figcaption></figcaption></figure>

* **Upload status feedback** — file uploads now show clear per-field status (Preparing, Storing, Preprocessing, Generating preview…) instead of byte-level progress bars.

<figure><img src="/files/OQmW75iODFoY7BhtXZnw" alt="" width="473"><figcaption></figcaption></figure>

* **Stop the agent mid-run** — a new Stop button lets you interrupt the in-app assistant during a turn.

<figure><img src="/files/K1UJRU4qkT1UggUfJgY3" alt="" width="240"><figcaption></figcaption></figure>


# Release 0.29.0

**Release Date:** June 2026

**Previous Version:** v0.28.0

***

### New nodes & capabilities

#### Pick Video Frame node

Extract a single still frame from a video as an image. Choose the frame either by **Position (%)** (0% = first frame, 100% = last, default 50%) or by **Exact frame number** (0-based). Out-of-range frame numbers are clamped to the nearest valid frame. Runs on CPU with no compute cost.

<figure><img src="/files/Bc0k9swmpDHAka3QGGM9" alt="" width="563"><figcaption></figcaption></figure>

#### Remove Video Background node

Isolates the subject of a video and outputs a transparent overlay.

<figure><img src="/files/IJmtJhoepmovmcodGib2" alt="" width="563"><figcaption></figcaption></figure>

#### Per-project LLM selector in the assistant panel

A new **Model** dropdown in the AI assistant (Agent Chat) panel lets you choose which LLM the assistant uses for that specific project. Defaults to the system setting until you pick one; unavailable models are disabled in the list.

<figure><img src="/files/QvUvFXv3Ri4EWvsnj27H" alt="" width="473"><figcaption></figcaption></figure>

#### Atlas Rigging nodes (for users with the `mesh_animations` role)

A new **Atlas Rigging** node group:

* **Atlas Rig T-pose Humanoid** — rigs a T-pose humanoid mesh and outputs a rigged GLB plus default animation previews.
* **Atlas Apply Mixamo Animation** — applies a Mixamo preset or a custom animation to a rigged mesh.
* **Atlas Mesh to FBX** — exports a rigged/animated mesh as a downloadable UE 5.5 FBX zip.

The rigging algorithm is selectable (labeled **v24.1rc3 GPU**).

<figure><img src="/files/2SDKlgoNKHQAfNPwlwSv" alt="" width="563"><figcaption></figcaption></figure>

***

### Models & node updates

#### Seedance 2 & Kling video updates

* **Seedance 2.0 Standard** now supports **4K** output (480p / 720p / 1080p / 4K); Seedance 2.0 Fast remains 480p / 720p. Up to 9 reference images.

<figure><img src="/files/Hg5DMo88i3l8Q322DOsL" alt="" width="527"><figcaption></figcaption></figure>

* **Kling Video 3.0** now offers **1080p or 4K** output, plus new **Kling O3 R2V Pro** (1080p) and **Kling O3 R2V 4K** variants.

<figure><img src="/files/sEXA9M1QgftdubcLnnX3" alt="" width="563"><figcaption></figcaption></figure>

#### Happy Horse 1.1

All Happy Horse video modes upgraded to **1.1**. Image-to-video input size limit raised from 10 MB to **20 MB**, lower 1080p pricing, and new aspect ratios (9:21, 5:4, 4:5, 21:9). Reference-to-video supports 1–9 references, 3–15 s, 720p / 1080p.

<figure><img src="/files/c4DcDmxvDcAmuSL8Wuu1" alt="" width="557"><figcaption></figcaption></figure>

#### Hyper3D Bang — new mesh parts-separation backend

The mesh parts-separation node gains a selectable **Hyper3D Bang** backend (alongside the existing Hunyuan3D Parts default). Requires the mesh plus a reference image (optional prompt) and outputs a segmented GLB. New controls: **Strength** (2–12, higher = more pieces), **Material** (PBR / Shaded / None / All), and **Texture Resolution** (Basic 2K / High 4K).

<figure><img src="/files/szmA4Wj7AuiCAh9fOZf9" alt="" width="563"><figcaption></figcaption></figure>

#### Contextual image-filter node controls

Selecting a filter on the image-filter node now reveals **only the parameters that filter actually uses**, each with a clear name (e.g. Noise Strength, Blur Amount, Opacity, Gamma, Radius). Blend-mode and color-transfer filters reveal a reference-image input only when relevant. Existing projects are migrated automatically.

<figure><img src="/files/YwoU7Z3RURWaoodDKS9s" alt="" width="563"><figcaption></figcaption></figure>

#### Crop / Paste image nodes reworked around region masks

Crop and paste nodes now exchange a **PNG region mask** (white rectangle on black) instead of numeric coordinates. A node builds a rectangular region mask from relative (0–1) coordinates that you connect to **Crop Image Auto** or **Paste Crop Into Image**. Crop padding modes (clamp / black / replicate) control behavior past the canvas edge. Existing projects are migrated automatically.

<figure><img src="/files/aIp0Dpb76qdpA8N8pGi5" alt="" width="563"><figcaption></figcaption></figure>

#### Mesh metadata stats

Mesh descriptions now include simple human-readable stats — e.g. *"12,345 vertices, 6,789 faces, triangulated topology"* — plus armature/normalized indicators when present.

<figure><img src="/files/pkxfsKfnFpDTXjoO3zLc" alt="" width="563"><figcaption></figcaption></figure>

#### Clone shared workspace projects via "anyone with link"

You can now clone a workspace project shared as "anyone with the link" into your own editable copy, even without being a member of that workspace — matching the read-only open behavior.

***

### Workspace & project changes

#### All projects are now workspace-based

"Personal" projects have been removed — every project now lives inside a workspace. The workspace switcher no longer shows a "Personal" entry (it reads "Select workspace" when none is selected), and creating a project with no workspace selected prompts you to add/select one first.


# Release 0.30.0

**Release Date:** July 2026

**Previous Version:** v0.29.0

***

### Projects

* **Duplicate a project** directly from the `…` menu on a project card. The copy lands in the same workspace (named "… (copy)") and appears in the grid without navigating away.

<figure><img src="/files/T6TvCSLgHfcbKAnUInhP" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/bL4lofGNQZEzG3yrf7r9" alt="" width="563"><figcaption></figcaption></figure>

### Accounts

* **Sign in with Apple** — a new Apple sign-in button alongside Google on the auth screen.

<figure><img src="/files/sJblYj32lyoMSgJIo0L8" alt="" width="526"><figcaption></figcaption></figure>

### Image & video generation

* **New Gemini image backend:** Gemini 3.1 Flash Lite Image for text-to-image and multimodal image composition.

<figure><img src="/files/OnieyBSWmc2AVpbCGBdb" alt="" width="563"><figcaption></figcaption></figure>

* **New Gemini video backends:** Gemini Omni Flash for Text+Image → Video, and **Gemini Omni Flash video editing** in the Video Edit node.

<figure><img src="/files/H1wUJPytTMIyjxuDL9fJ" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/UO0weKywWWEroc4cKF1K" alt="" width="563"><figcaption></figcaption></figure>

* **Stricter aspect-ratio / resolution control.** Image nodes replace the old 1K/2K/4K heuristic with explicit, backend-aware **Aspect / Resolution** modes (e.g. `1:1@2K`), so you get exactly the dimensions you pick.

<figure><img src="/files/Y7IygF48B7oneVNpQ1G7" alt="" width="563"><figcaption></figcaption></figure>

* **Advanced strength / guidance knobs** exposed for supported image engines (Hunyuan Image 3.0 guidance scale, FLUX.2 Flex guidance, Z-Image Turbo edit strength). These are advanced, engine-specific fields; defaults match previous behavior, so existing graphs are unchanged.

<figure><img src="/files/n7DPREvcqLAlCcFxUFqw" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/9NYKDK9cXXEONVoopnso" alt="" width="563"><figcaption></figcaption></figure>

* **Higher-detail 3D default:** Titan v1 default face count raised to 1M.

<figure><img src="/files/x9nxUCXmrRYw0EeiVX3i" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/JIaHjL119qaVMi8bmZ2W" alt="" width="563"><figcaption></figcaption></figure>

### Nodes & canvas editing

* **Multimodal image node refactor.** Image generation and editing are unified around the `Text + Image -> Image` node. Kling O1 is replaced by **Kling O3/V3** with conditional element/reference controls. Existing nodes are migrated automatically.
  * **Deprecation:** the standalone **Edit Image with Text** node is removed; saved edit nodes are migrated to `Text + Image -> Image` with prompt/image fields preserved.

<figure><img src="/files/dOkJjvd4FfTU2aLigBMe" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/NN5OLzisIRn9ilycctlj" alt="" width="461"><figcaption></figcaption></figure>

* **Interactive mask picking.** New **Ellipse Mask** node plus a visual mask picker for both rectangular and ellipse masks — drag/resize an overlay on the reference image, with live coordinate inputs. ("Region Manual" is renamed **Rect Mask**.)

<figure><img src="/files/Oce5tTFn4LuBQIpwhEFO" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/kLEzbT0DPJ4QKGkCkih5" alt="" width="563"><figcaption></figcaption></figure>

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fuu9ng52svFpSOi0pqVB3%2F20260710-0917-07.1094417.mp4?alt=media&token=50b50a8f-f5c4-430a-ac25-22679a217bc9>" %}

<figure><img src="/files/O0dQdAms6jsiAF8Dggaq" alt="" width="563"><figcaption></figcaption></figure>

* **Custom Image Format Output node.** Export images as **JPEG, TGA, WebP, TIFF, or AVIF** with per-format settings (quality, subsampling, compression, alpha handling, etc.). Warns when exporting from an already-lossy source.

<figure><img src="/files/4Xo50cKMCLLL85aGvWTx" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/jsxp9Ty9l4O0c0bnGOFZ" alt="" width="563"><figcaption></figcaption></figure>

* **Save long text outputs.** Multiline text outputs now have **Copy** and **Save as `.txt`** buttons in the node editor (download is disabled until the output finishes streaming).

<figure><img src="/files/BQJcGX43GeyJuB2L0ZHO" alt="" width="563"><figcaption></figcaption></figure>

* **New Invert Mask Node:** Flips a grayscale or binary mask on the pixel level, swapping editable and preserved regions. Useful for isolating the opposite of what a segmentation or background-removal model detects, without re-running prompts or hand-painting inverted shapes.

<figure><img src="/files/hWqBYhR6lmLR0C6zwD1P" alt="" width="563"><figcaption></figcaption></figure>

### Billing & workspace admin

* **Members can view their own billing.** Workspace members now reach the Billing page and see their credit badge, their own usage this period, and their own transaction history. Team-wide data and management controls (subscription, limits, plan, payment) remain owner/admin-only.

### API & developer changes

* **Unified prompt reference conventions.** Image/character references (`@image_1`, `@character_2`, and legacy aliases like `image_0`, `@image1`) are normalized deterministically and mapped to each provider's format, with a warning when non-canonical references are auto-resolved.

<figure><img src="/files/vyv8424KME5ppimoHcdJ" alt="" width="305"><figcaption></figcaption></figure>


# Release 0.31.0

**Release Date:** July 2026

**Previous Version:** v0.30.0

### Editor and workflow experience

* **Persistent navigation and project actions.** A new top bar stays in place across the dashboard and editor, with workspace or project context, project sharing and export actions, the credit balance, AI Assist, and the user menu close at hand.

<figure><img src="/files/GXyIfDnJ1LMSuvKr3qPO" alt="" width="563"><figcaption></figcaption></figure>

* **A new floating canvas toolbar.** Select, pan, section, comment, and note tools now share a compact modal toolbar with zoom-to-fit, zoom in/out, preset zoom levels, and a live zoom percentage.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F6N5ExO2gIaUU6rQS6gag%2Frelease%2031.mp4?alt=media&token=4a91590c-a759-402d-8d10-28bd0b50e531>" %}

* **Alignment and distribution shortcuts.** Selected nodes can be aligned left (`Shift+A`), horizontally centered (`Shift+H`), right (`Shift+D`), top (`Shift+W`), vertically centered (`Shift+V`), or bottom (`Shift+S`). Use `Ctrl/Cmd+Shift+H` or `Ctrl/Cmd+Shift+V` to distribute them horizontally or vertically.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FiazHupfCD1isBGaa2gVW%2F20260728-0811-23.1621208.mp4?alt=media&token=b005bf30-6937-4539-afe9-cafb9171c73a>" %}

* **A more compact node creation menu.** Icon chips, single-line results, and tooltips make nested node categories quicker to scan.

<figure><img src="/files/aGHBwpu6paXm1DQsCVfX" alt="" width="422"><figcaption></figcaption></figure>

<figure><img src="/files/gs6JKg5VvFzUujkWyVfV" alt="" width="419"><figcaption></figcaption></figure>

* **A redesigned AI Assistant.** The resizable side panel has a cleaner conversation view, starter prompts, visible tool progress, model selection, a Run nodes control, conversation clearing, and a Send/Stop control. The legacy one-click Suggestions panel has been retired so graph assistance is consolidated in the AI Assistant.

<figure><img src="/files/LGRLjgH6qPNLMeT7EFbO" alt="" width="392"><figcaption></figcaption></figure>

* **An upgraded file and 3D viewer.** The new pill toolbar provides download and camera controls, selectable lighting presets, and available material/debug views such as albedo, normals, ambient occlusion, roughness, metallic, part colors, and wireframe. Live information now includes image dimensions and file size, or 3D topology, vertex/triangle counts, physical dimensions, and texture resolutions.

<figure><img src="/files/8PETHSS4MthlrW9Ud3xI" alt="" width="563"><figcaption></figcaption></figure>

### Models and media tools

* **New Gemini choices.** OpenAI GPT 5.6 Sol/Terra/Luna, Anthropic Claude Sonnet 5, Google Gemini 3.6 Flash and Gemini 3.5 Flash Lite are available in the supported LLM/vision model catalog. Existing Gemini 3.5 Flash selections are migrated to Gemini 3.6 Flash.

<figure><img src="/files/D3zITGfO9cAfrs4ixO8t" alt="" width="467"><figcaption></figcaption></figure>

* **Tripo 8K texturing.** Direct Tripo 8K options are available for v3.1, P1, and Smart Low Poly generation in single-image and multiview workflows, plus Tripo 3.0 8K in Texture Mesh. These produce an 8K base-color texture with 4K material and NormalGL maps.

<figure><img src="/files/0fVlH7BOeEEx3pZW361K" alt=""><figcaption></figcaption></figure>

<figure><img src="/files/JUkyrV9z7bGf78Dfh1A3" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/6SPD5RlszoUsINdkHH3A" alt="" width="563"><figcaption></figcaption></figure>

* **More Meshy controls.** Image to 3D adds Meshy-T2 Smart Topology for clean single-image low-poly output with a configurable 100–15,000 face target. Meshy v6 workflows can generate 4K base-color textures in Image to 3D, Multiview to 3D, and Texture Mesh. UV Unfold also adds Meshy UV Unwrap as an engine for GLB meshes up to 40,000 faces.

<figure><img src="/files/p5D2jpwOpnSBHc4O6sVW" alt="" width="563"><figcaption></figcaption></figure>

### Atlas Platform MCP (alpha)

Early-access users can connect Atlas to Claude Code, Cursor, Codex, and other Model Context Protocol clients with a workspace API key. The initial toolset can manage, clone, and share projects and their team access; ask a one-shot Atlas agent to build or edit a graph; and export, list, rename, disable, or remove exported APIs. A guided MCP setup page is available in workspace settings.

The alpha MCP agent uses the same workspace permissions as the key owner. In this first version, its turns are stateless and it can construct graphs but cannot run nodes.

MCP alpha can be enabled upon request for now.

<figure><img src="/files/YwpWmo2t8J7ihXx2Hrpf" alt="" width="563"><figcaption></figcaption></figure>


# Release 0.32.0

Release Date: August 2026

Previous Version: v0.31.0

### Editor and workflow experience

* **A guided first-canvas tour.** New users can launch a five-step "Get started" video tour covering the toolbar, nodes, runs and results, Atlas AI, and exports. The tour is also available from the mobile presentation-only canvas.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FZxCo9Th6QG4nDJ3N4S0l%2F20260813-1121-13.3920162.mp4?alt=media&token=9cc77b8b-da41-428e-9658-1d1728067e01>" %}

* **Interactive Atlas AI node references.** Node references in Atlas AI responses render as named chips; selecting one highlights and centers the referenced node on the canvas.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FfWp7ujGjTdztiA9eVigI%2F20260813-1123-57.2680499.mp4?alt=media&token=2c87af3b-8cfc-4da9-86a2-d626ea8284e5>" %}

* **A cleaner full-screen file viewer.** The redundant file-type badge has been removed from the viewer.

<figure><img src="/files/mddL222zHd1khuFtPLWM" alt="" width="563"><figcaption></figcaption></figure>

### Models and media tools

* **Expanded image and video model lineup.** Added Seedream 5 and Seedream 5 Pro for image generation and editing, including workflows with up to ten reference images. Added Seedance 2.0 Mini and Seedance 2.5 Pro for text-, image-, and reference-to-video workflows. Seedance 2.5 Pro also supports Video Edit and Video Extend, 4–30 second clips, optional audio generation, and image, video, or audio references. Eligible Seedream and Seedance workflows can also use non-celebrity face references after explicit confirmation.

<figure><img src="/files/nik1Z9RvFGQruyU74n8J" alt="" width="563"><figcaption></figcaption></figure>

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F4nhqeabnTBYbfUfmlds6%2Fseedance2_reference.mp4?alt=media&token=134a876a-8a42-4210-a32c-10bd2feefee0>" %}
Seedance 2.5 Pro Reference to Video
{% endembed %}

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F6a4bOotegagVAiaoC8im%2Fextended_video.mp4?alt=media&token=80509547-0a1c-4909-8ba2-26639adc5093>" %}
Seedance 2.5 Pro Video Extend
{% endembed %}

<figure><img src="/files/jDZRjxIbfcRCwPmmbw73" alt="" width="563"><figcaption></figcaption></figure>

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fo9EnGAXuSWmaRtWOPrcC%2Fgenerated_video.mp4?alt=media&token=2f6a632a-8981-46d4-ae66-ecdba7ef2d36>" %}
Seedance 2.0 Mini Text + Image to Video
{% endembed %}

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fn6wMkaRijOMl9MrpHxI4%2Fgenerated_video%20(1).mp4?alt=media&token=4a7a0a1b-af1b-41a6-b1ca-ff47a433d706>" %}
Seedance 2.5 Pro Text + Image to Video
{% endembed %}

* **Six additional BytePlus language models.** DeepSeek V4 Flash GA, DeepSeek V4 Flash, DeepSeek V4 Pro, Seed 2.1 Turbo, Seed 2.0 Pro, and GLM 5.2 are available in applicable text, structured-output, vision, and agentic nodes.

<figure><img src="/files/MHmHG3Zv8L9eapQSsLU4" alt="" width="563"><figcaption></figcaption></figure>

* **More Atlas Rig controls.** Where Atlas Rig is enabled, the T-pose Humanoid node now offers Procedural, AI, AI\_2, and combined hand processing, an optional rotation-repair pass, and surfaced worker warnings.

<figure><img src="/files/bHplDVCB6bbllrZeuHe6" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/CALcQK1MtpOLMGSuBWPz" alt="" width="563"><figcaption></figcaption></figure>

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FPDFpiEQVFcbFjAomjYGQ%2Fatlas%20rig.mp4?alt=media&token=370997d2-ca66-4f6e-9095-1f8d851265e9>" %}

### Image options

* **Exact input-image dimensions.** Image workflows with source images now offer an "Exactly match input image" mode. Atlas pads the source to a provider-supported canvas and crops the result back to the first input image's exact pixel dimensions; the existing approximate mode remains available as "Roughly match input image."

<figure><img src="/files/L1O5nEFolSf6U6HBIgW8" alt="" width="563"><figcaption></figcaption></figure>

* **Drag-and-drop image ordering.** Images in multi-image node fields can be reordered with a mouse or touch long-press, including automatic scrolling for longer lists. Reordering is safely disabled while files are uploading.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FjRKmYSfzKpi2Gmekkf8L%2F20260813-1302-01.9483969.mp4?alt=media&token=11056910-57ef-4f5e-aae7-03a1db6030f8>" %}

### Account and sign-in

* **Simpler login and signup.** Google and Apple sign-in are now presented first. The blocking Terms checkbox has been replaced by a linked "By continuing…" notice.

<figure><img src="/files/79O4Ky75TvHeA5dWhFVk" alt="" width="501"><figcaption></figcaption></figure>

<figure><img src="/files/fBzldhIewbxoZjp5jy2U" alt="" width="484"><figcaption></figcaption></figure>

* **A one-time onboarding email.** Newly verified users receive a welcome email with getting-started guidance, independently of whether trial credits are enabled.

<figure><img src="/files/hZRPsLkh9M5Fv04dpFTy" alt="" width="471"><figcaption></figcaption></figure>

* **Mobile-to-desktop continuation.** Mobile users who open an editable project can receive a one-time email link for continuing that project on desktop, with confirmation shown in the mobile guidance.

<figure><img src="/files/WOc252oeKwZXc5auSlXy" alt="" width="563"><figcaption></figcaption></figure>

<figure><img src="/files/ntPpSECKZzr6q4fEHNAh" alt="" width="563"><figcaption></figcaption></figure>

* **Magic-link sign-in retired.** The legacy magic-link sign-in flow and `/api/magic-link` endpoint have been removed; email/password and supported social sign-in methods remain available.

### Atlas Platform MCP (alpha)

* **MCP project provenance.** Projects created through MCP are labeled "via MCP." MCP `list_projects` and `get_project` responses expose `createdViaMcp`, and users with MCP experimental access can filter their Projects page by MCP origin.
* **Broader Atlas AI and MCP capability awareness.** Atlas AI and MCP client guidance now covers the platform's image, 3D, video, audio, document, and workflow capabilities more comprehensively when answering open-ended questions.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F83AlxwfeoNJNBQLYPNQf%2Fmcp.mp4?alt=media&token=8ae2f278-75ac-4f42-a9d7-b8f9454c4454>" %}

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FFPROxrqRBIXU5p01EnHT%2F20260813-1116-23.8628858.mp4?alt=media&token=6fc00f31-d47b-485c-93c8-4f47efd0f45c>" %}