Release 0.42.0
Previous Version: v0.41.0
Choose how deeply supported AI models reason, add suggested nodes from the assistant, inspect how an asset was generated, and organize canvas notes through the assistant. This release also adds 3D engine options, simplifies model choices, and makes credit-loading states explicit. These are the new features for normal users since 0.41.0, including the updates in 0.41.1. Workspace owners and admins are included; available nodes and options depend on account and workspace access.
Choose reasoning effort for supported AI models
Select the reasoning effort alongside the model in Text Generation, Structured Output, Image to Text, Images to Text, Landmark Selection, Procedural Image Builder, Composition Plan, and Blender MCP nodes. The project assistant also gains a compact effort selector beside its model selector.
Choices adapt to the selected model, such as Low, Medium, High, or Max where supported. Models without supported named levels hide the control.
Choices are saved with the node or project assistant and survive reloads. Switching models keeps a compatible level and resets an incompatible visible level to Default.
Default preserves existing provider behavior, including Landmark Selection’s existing Low default for Gemini. Explicit choices are passed to the selected provider, including grounded text, structured-output, and caption requests.
The shared node-input contract exposes
reasoning_effort; API-driven and assistant-created workflows use the same model-specific validation. Procedural Image Builder also records the effort in generated PNG metadata.


Add suggested nodes directly from assistant replies
Exact node-type codes in assistant replies, such as image_to_3d, now appear as named add buttons with a plus icon and the node’s variant tag. Clicking a button, or activating it with Enter or Space, creates that node with its default settings beside the graph, selects it, and brings it into view. Each addition is one undoable edit.
References to existing nodes continue to select and center those nodes. Visible but unavailable node types are marked as unavailable; read-only projects show names without add buttons. The assistant is instructed to distinguish suggested node types from field codes and existing-node references.
Inspect the recorded generation behind an asset
The asset detail card now shows the recorded run that produced an asset, alongside separate Origin, Project, Created by, Created, and Source node fields.
See the recorded prompt with a Copy action, the model or engine, and a compact settings summary. +N more expands additional and advanced settings. Only settings belonging to the recorded engine appear; recorded LoRA settings include paths and strengths.
Inspect up to eight recorded input references, with names and provenance on hover and a count for additional inputs. Repeated inputs retain their recorded order. Missing originals are marked Deleted; references unavailable through the workspace catalog are marked Unavailable. Unrecorded provenance stays Not recorded.
Access Add on canvas, Rebuild, Rebuild chain, Download, and a direct Open in viewer button from the card. Rebuild actions continue to restore the recorded workflow without executing nodes; unavailable rebuilds now explain the reason inline and in tooltips, and placement displays Rebuilding….
Uploaded assets show source attribution without generation or rebuild controls. Generation details depend on the metadata actually recorded for the asset.

Ask the assistant to read, write, and organize canvas notes
The project assistant can now use the full text of canvas notes as context and create, edit, resize, restyle, reassign, detach, and delete notes. This capability is also available through MCP’s existing prompt_project_agent workflow.
Notes can belong to a node or frame, use inferred ownership, or remain explicitly detached. Inference favors the containing frame, then the nearest node or frame.
During assistant arrangement, attached notes travel with their owner. Notes attached to a node retain their relative position; notes attached to a frame are organized with its contents. Detached notes stay independent. These explicit attachments govern assistant operations; they do not add automatic note-following to manual node dragging.
When removing a frame, the assistant reattaches its notes to a surviving node inside that frame where possible. Otherwise it must reassign, detach, or delete them. Deleting a node with attached notes requires the same decision before deletion proceeds.
Saved notes retain their text and ownership across reloads. Copying nodes or frames with their notes remaps explicit note ownership to the copies.
The assistant’s graph-grouping response now includes note text, geometry, ownership, and attachment mode. Its existing arrangement operation accepts note edits and reports updated and deleted notes; these are assistant capabilities, not new standalone external MCP tools.
New 3D options and clearer engine controls
Multiview to 3D adds Meshy v7.1 Quad, alongside the existing triangle option.


Retexture Mesh adds Tripo 3.5 8K, including De-light. Tripo 3.0 remains as an explicitly labeled compatibility option for geometry from Tripo 3.0/3.1.
Titan gets its own Auto/Custom face-count controls, independent of other engines in Image to 3D fallback chains. Choose Auto (1,000,000 faces) or a Custom target from 20,000 to 2,000,000 faces. Saved targets and connections are migrated; saved numeric targets outside Titan’s supported range are clamped. A connected face count below Titan’s 20,000-face minimum no longer prevents an eligible earlier engine from running: Titan is skipped if that attempt is reached.

Model descriptions and assistant guidance recommend Meshy v7.1 for generation and v7 for retexturing, while retaining v6 for specific capabilities such as requested emission maps or v6-only De-light, Lite generation for budget requests, and specialized topology modes.
Consolidated model choices, with saved-workflow upgrades
Duplicate and superseded choices are removed from the model pickers. Saved projects and exported APIs upgrade to the replacements through the shared project migrations.
Seedream 5 / Standard
Seedream 5 Lite
General Gemini 3.1 Flash Image
Nano Banana 2.1
Kling V3 image
Kling O3
Meshy 6 Lite retexturing
Meshy v7
Tripo 2.5 retexturing
Tripo 3.5
Tripo 3.0 8K retexturing
Tripo 3.5 8K
Legacy Grok text/image-to-video
Grok Imagine Video 1.5 Lite
Legacy Grok reference-to-video
Full Grok Imagine Video 1.5, preserving reference inputs
Seedance 2.0 Fast, including reference workflows
Seedance 2.0 Mini
Seed 2.0 Pro and Code
Seed 2.1 Turbo
Gemini 3.1 Flash TTS preview
Gemini 3.8 Flash-Lite TTS
Eleven v3 single-speaker TTS and Eleven Multilingual v2
Eleven v4
Add Audio Tags: legacy Gemini target
Gemini 3.8 target
Relevant saved controls, connections, paid outputs, and run history are retained. Changed requests invalidate their execution caches; unchanged single-Titan requests retain their paid cache entries. Reruns use the replacement model’s behavior and pricing. In particular, the full Grok 1.5 reference route has a higher price than the retired route.
Additional compatibility and availability changes:
Speech: migrated Gemini TTS nodes translate legacy bracketed vocal events and move delivery/emotion directions into Style, including text from connected generators and exported API parameters. Source text and upstream cached results are preserved. New Gemini 3.8 nodes retain verbatim text semantics. Add Audio Tags defaults to Eleven v4 for speech; Eleven v3 remains available for dialogue, including compatible saved dialogue taggers. Eleven Flash v2.5 remains available for long single-speaker text.
Music: valid legacy
sectionscomposition plans convert to Music 2.5chunks, including plans supplied by connected generators or exported APIs. Lyrics, styles, durations, and source-audio ranges are carried over. Plans using excluded source ranges retain the Music v1 path because that control has no equivalent in Music 2.5. Existing paid outputs remain available; explicit reruns use the appropriate current rendering path.Legacy FLUX training: new legacy fine-tune training is disabled. Completed training records become read-only records; their saved fine-tune IDs and inference remain usable. Downstream inference can reuse those records without waiting for old training-image dependencies to run.
Retained choices: 404-GEN remains available alongside Titan. Meshy Lite generation, specialized topology modes, Gemini 3.1 Flash-Lite text, and supported dialogue options remain available for their distinct uses.
Veo: the preview Developer API routes are removed from generation fallback chains; the existing Vertex, fal, and Replicate routes remain.

Explicit credit-loading states and a workspace billing query
Credit displays now distinguish Loading credits…, Refreshing…, Credits unavailable, and out-of-date observations. When a refresh fails, the last observed balance stays visible as stale while the client retries. Paid runs and affected billing controls wait for usable billing information.
The authenticated Socket.IO query get_workspace_billing returns the current member’s balances, usage, limits, payment-lock state, and billing metadata together. Workspace owners and admins receive the team’s configured member caps; other members receive their own cap information. The query requires current membership in an active workspace and excludes billing histories and provider-internal records. This is a session-authenticated socket API, not a new API-key REST or MCP endpoint.


Run the nodes inside a canvas frame
In an editable project, click a frame’s header to access run controls for its nodes, including nodes inside nested or collapsed frames. Combining selected frames and nodes counts each node once. The toolbar summarizes ready, up-to-date, and blocked work; empty frames show Nothing to run.
Run N nodes submits ready nodes. Run with prerequisites also schedules the required connected inputs. Cancel queued cancels queued work and skips its queued dependents while upstream work continues. Existing access and billing restrictions still apply.
The selection’s credit estimate appears when available. Linked nodes within the selection can show Cost unavailable, and prerequisite work can add to the selection estimate.


Jump directly to a missing input or connected source
Click a blocked node’s warning badge to see what needs attention. Show input locates and focuses the affected field, revealing Advanced settings when needed. Inline guidance clears as you correct the inputs.
Show source opens the connected source node and unfolds enclosing frames. These actions locate the problem without running or scheduling the workflow. Unavailable sources use neutral wording without a navigation action.



See the current shortcut beside canvas actions
Tooltips for Section and Note now show their current shortcuts, including saved custom bindings. The project menu also shows the Export API shortcut where available. Hints update with your bindings, use platform-appropriate key labels, and omit unassigned shortcuts.
Clicking Section or Note starts drawing on the canvas; the corresponding shortcut adds an item immediately.

Last updated