> For the complete documentation index, see [llms.txt](https://docs.atlas.design/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.atlas.design/atlas-ai-studio-overview/release-notes/release-0.50.0.md).

# Release 0.50.0

### Find the right kind of Lab

Atlas Labs now separates **Quick Labs** from **Lab workspaces**. Quick Labs take one input to one result, including tools that also run on your phone. Lab workspaces provide their own interfaces for multi-step work and open on prepared examples. The featured Lab stays above both shelves.

For game development, use a Quick Lab for a single prop repaint, then explore a Lab workspace for a larger environment or character workflow. Available Labs depend on the workspace catalog; this change organizes the desktop catalog and does not add a new mobile runner.

![Quick Labs catalog](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FiPSpsmvlVvxLk5zZfcGf%2Flabs-quick.jpeg?alt=media)

![Lab workspaces shelf](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FPJJyLLaW9PKmMYwzCMZV%2Flabs-workspaces.jpeg?alt=media)

**Game-development walkthrough.** Open Repaint's free prepared prop example, then choose **Open as graph** to inspect a private workflow copy. This uses the shipped example; no provider generation was started. Open Landscape Creator from the workspace shelf to explore the prepared Skull Island terrain example and its Unreal R16 elevation controls.

![Free Repaint example](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FKXHKVNCNV7M5pvSTyiGv%2Frepaint-free-example.jpeg?alt=media)

![Repaint private four-node handoff](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FXzgV3NdjAJAq07D2ctKZ%2Frepaint-private-graph.jpeg?alt=media)

![Landscape Creator prepared terrain](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FOUrjpCl7jdvRBRsFEvHe%2Flandscape-prepared-example.jpeg?alt=media)

### Prototype cinematics with P-Video-2

**Text + image to video** adds **P-Video-2 Standard** and **P-Video-2 Pro / MiniMax H3**.

* **Standard:** generate 1–20-second clips or let the model choose the length. Choose 720p or 1080p, 24 or 48 fps, and draft mode for early previews. Supply optional first/end frames or a soundtrack; the soundtrack determines clip length.
* **Pro:** generate 5–15-second clips with Cost, Speed or Quality mode. Choose native 480p/768p or upscaled 1080p, 2K or 4K output. Pro keeps 24 fps and generated audio.

<figure><img src="https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F5BfFj4LBwHzqgh85RWeF%2Fimage.png?alt=media&amp;token=26cfd3ec-8ec0-416d-b647-aebc2e2abdf7" alt=""><figcaption></figcaption></figure>

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FdhnBlu84Oq79XCyAXl3n%2Fp_video_2.mp4?alt=media&token=718411e6-5aa4-43c1-b078-6be9ca440ff1>" %}
P-Video-2 Pro
{% endembed %}

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FNPpK40aKoxFUMCFutljz%2Fp_video_2%20(1).mp4?alt=media&token=e36f5c84-4d4f-42f3-906d-5d29846fb746>" %}
P-Video-2 Standard (Draft Mode)
{% endembed %}

### Send the next instruction while the assistant works

You can now submit follow-up messages while the sidebar assistant is busy. Atlas queues them in order, then sends them together after the current tool finishes. Work already started on nodes continues, and node-completion notifications remain active.

For example, ask for an RPG quest outline, then queue “Make the quest non-combat” and “Add an optional puzzle reward” while the assistant builds the workflow. You can refine the brief without waiting for the whole turn to end.

The queue belongs to the current browser tab and clears on reload, project/workspace change or version checkout. **Stop** clears queued messages and subscriptions; it does not cancel node work already started. If delivery becomes uncertain after a connection loss, the queue pauses for an explicit **Retry** because a repeated message can repeat work.

![Queued game-design follow-up](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F2FKPekez9zEmMzpIgfEX%2Fqueued-followup.jpeg?alt=media)

### Choose GPT-6.1 Terra for text, vision and tool workflows

**OpenAI GPT-6.1 Terra** joins the shared model choices used by text, image-to-text, structured-output and agentic tool-use nodes. Reasoning controls range from **Off** through **Max**, with **Medium** as the default.

Existing model choices remain available. Workspace permissions and provider availability still govern execution.

<figure><img src="https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FABBwuAGCOpx9mdnoZL3d%2Fimage.png?alt=media&amp;token=b620e565-51ad-4329-9634-457f3e4840df" alt=""><figcaption></figcaption></figure>

### Turn recorded dialogue into timed text with Kotoba

**Speech to text** adds **Kotoba** for recorded clips up to two minutes. Select the spoken language explicitly: Japanese, English, Korean, Mandarin, Taiwanese Mandarin, Hong Kong Cantonese or Spanish. Optional controls provide word timestamps, speaker identification, vocabulary hints and Japanese katakana-name handling.

<figure><img src="https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fz5hP6VFkXkUYKUqo4nTV%2Fimage.png?alt=media&amp;token=55c83c06-f8b3-4bc8-b861-87ec61f75be6" alt="" width="354"><figcaption></figcaption></figure>

<figure><img src="https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F74kvGZaAc5kW5Hg4eQds%2Fimage.png?alt=media&amp;token=2a109cf5-c0e9-4ab1-a69e-dd83c729180e" alt=""><figcaption><p>Example Workflow that could be used for timed subtitles</p></figcaption></figure>

This addition handles recorded transcription. It does not include live microphone capture, streaming or translation. The existing strict IP-safe policy can restrict Kotoba.

### Replace a performance with your game character

**MiniMax H3 Character Swap** replaces the main subject of a 2–15-second guide video with a character from a reference image. The guide supplies the scene, camera, motion and timing. Keep its existing soundtrack, or request H3-generated audio. The result is trimmed to the guide length; resolution can match the guide or use 480P, 768P, 2K or 4K.

For a game cinematic, apply an original character design to a short performance reference before committing to a full animation pass. The reference character is isolated onto neutral gray before generation. **Keep guide audio** requires an audio track in the guide. Workspace permissions and IP-safety policy apply. The merged change identifies licensing review and the deployment's private LoRA mirror as rollout prerequisites; broad availability is not asserted here.

<figure><img src="https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FI8khQWak9jSjYPOi8wt7%2Fimage.png?alt=media&amp;token=02d98d04-c185-43d2-a37c-3c8b3b2a7461" alt=""><figcaption></figcaption></figure>

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F2E5sh59bDc2IlVFXv9iy%2Fgenerated_video%20(15).mp4?alt=media&token=9f272969-f8a7-4922-b3f4-6b68d2432be5>" %}
Original Video
{% endembed %}

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2For4xRPtc4XiuUOy6NqNW%2Fh3_character_swap.mp4?alt=media&token=95246c48-accb-431a-a395-ba9a1472ae75>" %}
Character Swap
{% endembed %}

### Generate and edit longer shots with Gemini Omni 1.2 Flash

**Text + image to video** now runs **Gemini Omni 1.2 Flash** for its Gemini Omni engine. Clips run 3–15 seconds (8 by default) at 360p, 720p, 1080p or 4K in 16:9 or 9:16, with generated audio. Prompts can bind up to ten reference images as @image\_1 to @image\_10; with a single image, Omni animates that image. Per-second prices are lower than for Omni 1.1. If Google cannot take a request, Atlas tries fal.ai and then, for prompts with at most one image, OpenRouter, and charges at the rate of the provider that served the clip.

**Edit video** adds **Gemini Omni 1.2 Flash Edit** alongside the unchanged Omni 1.1 edit engine. It applies a plain-language instruction to source clips of up to 15 seconds at 360p, 720p, 1080p or 4K. Longer sources are trimmed to just under 15 seconds with a warning, and the edit is charged per second of the trimmed source. Simple instructions work best; end with “Keep everything else the same.” to protect the rest of the shot. Voice editing and reference images are not supported.

Existing Omni nodes keep their settings. A run whose prompt, images and settings exactly match an earlier Omni 1.1 run can return that cached clip instead of generating with 1.2; change an input to get a 1.2 result. Workspaces that route Omni through their own GCP Enterprise Connection use Atlas credits until **Gemini Omni 1.2 Flash** is switched on there again or Atlas migrates the saved choice. Omni edits always use Atlas credits.

![Stormwatch level intro: generate, then edit](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FC50uelr16AySg2H3DPjy%2Fomni12-stormwatch-graph-720p.jpeg?alt=media)

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FaafPsVuycGubKLHLAx5J%2Fomni12-stormwatch-storm-original.mp4?alt=media>" %}

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F9AJyfG3yCSbW6DMJi2If%2Fomni12-stormwatch-winter-edit-original.mp4?alt=media>" %}

### Animate concept art quickly with Kling 4.0 Flash

**Text + image to video (Simple)** adds **Kling 4.0 Flash 720p**: 3–20-second 720p clips with native audio from a prompt or a first frame. Text-only shots can use 16:9, 9:16, 1:1 or 21:9; with a first frame, framing follows the image, optionally outpainted. There is no end frame.

**Edit video** adds **Kling 4.0 Flash** with **Edit source** and **Generate from reference** operations. Both take a 3–20-second MP4, MOV or WebM source of at least 700 pixels per side at 24–60 fps, accept up to four reference images and four Kling elements, and return a 720p clip of 3–20 seconds or the source length. Edits are billed on the source length as well as the output.

Kling 4.0 Flash costs less per second than Kling 4.0 and runs on a single provider, without a fallback. The Duration list is shared with Kling 4.0, so choices above 20 seconds are rejected before generation, without a charge. The existing strict IP-safe policy can restrict these engines.

![Ember Forge: concept frame, animation and blue-flame edit](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FajhrrjHlLA2xkCZ3cdup%2Fkling4flash-emberforge-graph-1080p.jpeg?alt=media)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.atlas.design/atlas-ai-studio-overview/release-notes/release-0.50.0.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
