Release 0.41.0
Previous Version: v0.40.0
Organize large workflows with collapsible frames, navigate the canvas from assistant replies, reserve credits using your connected media, and attach 3D meshes to a surface. These are the new features for normal users since 0.40.0; workspace owners and admins are included. Available nodes and provider options depend on your account and workspace access.
Collapse, move, and copy complete canvas frames
Collapse a group or section into a compact title strip to keep a large workflow readable. The strip shows the total number of nodes in its nested contents. Connections crossing the frame remain visible, while internal connections and contained nodes, notes, and frames are hidden. Expanding restores the contents, including any nested frames that were already collapsed. Folding changes the view without changing how the workflow runs.


Frame membership is saved with the project. Copying a selected frame includes its complete nested contents—even hidden items—and internal connections. Pasted frames start expanded, keep their relative layout and grouping, and remain selected with their nodes so you can move the copy together. Overlapping selections copy each item once.
Drag a frame by its header or empty tinted interior to move its entire subtree. Frames retain resize handles and grow to fit their children. Shift-drag through the empty interior starts a box selection; Space-drag pans; right-click opens the add-node menu.
Drop a node, note, or frame into another frame to change its membership, or drag a nested frame out to remove the nesting. Removing a frame keeps its contents and moves them into its surviving parent; Undo restores the grouping.
Locating a hidden node through the assistant or asset picker opens its enclosing frames. Read-only embeds preserve saved folds and hierarchy, and the mobile node list keeps all nodes visible.
Add to selections and constrain movement with Shift
Shift-click adds frames and notes to your existing selection, including a selection containing nodes. Command-click and Control-click also add these annotations. Nodes retain their configurable Add to selection shortcut, which defaults to Shift.
Hold Shift while dragging a node, frame, or note to constrain movement to the dominant horizontal or vertical axis. Mixed selections move together, and selected frames carry their nested contents, including collapsed items. Placement is saved and supports Undo / Redo.
Navigate to frames directly from assistant replies
Assistant references to groups and sections now show the frame’s name and a frame icon. Click a reference, or activate it with the keyboard, to select and center the frame. Enclosing collapsed frames open first; a target frame that is itself collapsed stays collapsed.
Names follow frame renames, and references to deleted frames return to plain text. Saved conversations also resolve these references. Read-only views show the names without interactive controls.

Reserve credits using your connected media
Supported media-processing nodes now calculate their upfront credit reservation from the connected media’s measured size or duration. Shorter audio and video can need fewer reserved credits. Supported image upscaling uses the source aspect ratio and requested output dimensions instead of assuming a square output.
Image size



Supported routes include Video Upscale, Lipsync, Video Edit, Genjutsu, video background removal, Voice Changer, Seed ASR, ElevenLabs Scribe, Mirelo music modification, and SeedVR2 / CGI image upscaling. Eligibility depends on the chosen provider and settings.
The canvas’s Up to … prices, selection totals, assistant maximum-credit estimates, and run admission use the stored measurements. New uploads and generated files record them; older files can be measured when a run starts. Missing measurements or a failed probe keep the conservative maximum, and some routes include a small reservation margin.
This changes how many credits are reserved before a run. It does not introduce a blanket reduction in generation prices.
Audio duration and settings

Credit rounding can make different short clips display the same reservation.

Video duration and routes


Predicted credit holds for eligible nodes
Nodes with maximum-based pricing can use a hold predicted from historical run costs when a usable trained model supports the current inputs. A smaller hold leaves more credits available for other runs while work is in progress.
The node shows Hold N credits, or Next hold N credits while running. The selection toolbar also uses Hold when its total includes statistical reservations. Quotes refresh after relevant input, media, permission, and pricing changes, and periodically while the editor is open.
The admitted hold stays fixed for that run. Final node charges cannot exceed it: Atlas absorbs costs above the hold, and refunds the unused reservation when a run costs less. Later estimates describe a subsequent run.
Predictions use execution inputs and measured media properties and are shared across users within the same platform environment. Nodes without sufficient evidence or a compatible model keep their formula-based maximum; exact-priced nodes keep their existing charging behavior. The assistant’s estimate_node_credits tool continues to return a deterministic maximum.
Lower predicted holds depend on trained models being available and are not guaranteed for a particular node or a fixed percentage.
Attach a 3D mesh to a surface
The new Attach mesh to surface node identifies a plausible flat mounting or support region and turns it toward a selected axis. Choose downward −Z (the default), upward +Z, forward −Y, backward +Y, left −X, or right +X.

Scale to Attachment Area uniformly scales the mesh to a requested contact area in square scene units. It starts enabled; turn it off to retain the original scale. The contact-area measurement excludes holes and gaps between separate feet.
Move Origin to Attachment Center places the chosen region’s center at the origin. It also starts enabled and can be controlled independently of scaling. When off, rotation and scaling use the existing origin.
Outputs include the transformed mesh, an Attachment Preview marking the chosen region on the original asset, and a Selection Reason.



The node supports static GLB meshes and examines their six axis-aligned sides. A mesh without a suitable flat region returns an explanatory error. Prepare animation, rigging, morph targets, GPU instances, and sparse or normalized integer geometry before using attachment.
Last updated