The canvas was getting corrupted to a small strip because of confusion
between two different dimension types:
- Output area dimensions (logical working area, e.g. 512x512)
- Display canvas dimensions (actual pixels shown on screen)
Root cause: Setting canvas.width/height attributes to match output area
while also using CSS width:100%/height:100% created conflicts. When
zooming or reloading, wrong dimensions would be read and saved.
Fix: Remove canvas element width/height attribute assignments. Let the
render loop control display size based on clientWidth/clientHeight.
Keep output area dimensions separate.
This prevents the canvas from being saved with corrupted tiny dimensions
and fixes the issue where canvas would only show in a small strip after
zooming or reloading workflows.
Enhanced LayerForge input handling to strictly separate image and mask loading based on connection type. Images now only load when allowImage=true and masks only when allowMask=true, preventing unintended cross-loading between input types.
Implemented dynamic brush cursor with visual feedback for size (circle radius), strength (opacity), and hardness (solid/dashed border with gradient). Added overlay canvas system for smooth cursor updates without affecting main rendering performance.
Redesigned the positioning system for the Blend Mode menu, inspired by the "Custom Output Area" logic. The menu now anchors precisely to the top-right corner of the viewport and stays in place during panning and zooming.
Refactored the "Output Area Size" button to use the new setOutputAreaSize method:
Correctly sets output area size and position based on current system logic (custom shape, extensions, outputAreaBounds)
Fully functional across all editor modes
New resizing logic:
Operates relative to the current center of the output area
Output area expands or contracts around its center without repositioning
Ensures center remains unchanged as expected
This fix provides precise control over layout dimensions and aligns with user expectations across workflows.
Introduced a drawTextWithBackground helper to centralize and simplify text rendering with backgrounds in CanvasRenderer (JS/TS). Updated all relevant usages to use this helper. Adjusted initial viewport x/y offsets in Canvas and Canvas.ts for improved default positioning.
Refactors the Canvas class to rename the 'canvasMask' property to 'maskEditorIntegration' for consistency. Updates all references and import to use the new name with the .js extension.
Renamed the CanvasMask class and file to MaskEditorIntegration for improved clarity and consistency. Updated all references in Canvas and SAMDetectorIntegration to use the new name.
Cleaned up the codebase by removing unused or redundant methods from Canvas, CanvasIO, CanvasLayers, CanvasLayersPanel, CanvasRenderer, and MaskTool. This reduces code complexity and improves maintainability without affecting core functionality.
Refactors how custom output area shapes interact with extensions, ensuring the shape's position and mask application remain consistent when extensions are toggled. Moves output area shape logic to CanvasInteractions, tracks original shape position, and updates all rendering, IO, and mask operations to use the correct coordinates. Improves mask chunk clearing and adds chunked mask application/removal for shape masks, ensuring correct behavior with expansion and feathering.
Output area bounds are now positioned relative to the world, not always at (0,0), and are updated to match custom shape placement. Rendering and extension logic have been updated to respect the new bounds, and the mask tool now adjusts to the output area position. Also sets log level to DEBUG for development.
This update introduces a unified output area bounds system, allowing the output area to be extended in all directions independently of the custom shape. All mask and layer operations now reference outputAreaBounds, ensuring correct alignment and rendering. The mask tool, mask editor, and export logic have been refactored to use these bounds, and a new UI for output area extension with live preview and tooltips has been added. The code also improves logging and visualization of mask and output area boundaries.
Introduces a CustomShapeMenu UI component for managing custom output area shape masks, including options for auto-applying the mask, expansion/contraction, and feathering. Updates Canvas and MaskTool to support these new mask operations, and ensures the menu is shown or hidden based on shape presence. Adds distance transform-based algorithms for accurate mask expansion and feathering.
Introduces a 'visible' property to layers and updates all relevant logic to support toggling layer visibility. Adds visibility toggle icons to the layers panel using a new IconLoader utility, with SVG and fallback canvas icons. Updates rendering, selection, and batch preview logic to respect layer visibility. Also improves blend mode menu UI and ensures new/pasted layers are always added on top with correct z-index.
Introduces ShapeTool to allow users to define custom polygonal output areas by holding Shift+S and clicking to add points. The selected shape is used to crop and mask images and layers, and is visualized on the canvas. Updates Canvas, CanvasIO, CanvasInteractions, CanvasLayers, CanvasRenderer, and types to support shape-based output areas, including shape-aware import, export, and rendering logic.
Adds the willReadFrequently: true option to all getContext('2d') calls on temporary canvases in Canvas, CanvasLayers, and SAMDetectorIntegration modules. This improves performance for frequent pixel data reads, addressing potential browser optimization issues.
Replaces setPreviewVisibility(false) with direct assignment to previewVisible in Canvas and Canvas.ts. Adds initialization of preview state based on widget value in CanvasView and CanvasView.ts to ensure correct preview visibility on widget creation.
Replaces the local auto-refresh toggle with a node widget property 'auto_refresh_after_generation' in both JS and TS Canvas classes. Updates Python CanvasNode to include this property in the required config and removes 'hidden' from 'trigger' and 'node_id'. This change centralizes auto-refresh state in the node widget for better consistency and UI integration.
Introduces a 'generationArea' context for batch image generation, ensuring that batch preview outlines and image placement remain accurate when the canvas is moved or resized. Updates related logic in Canvas, CanvasInteractions, CanvasLayers, and CanvasRenderer to track and render the correct area, and synchronizes context updates across user interactions.
Refactored batch preview management to allow multiple BatchPreviewManager instances per canvas. Updated positioning logic to use an initial spawn position, adjusted UI updates, and ensured batch preview menus move correctly with canvas panning. Removed single-instance references and updated related event handling.
Added draggable functionality to the batch preview menu, allowing users to reposition it within the canvas using world coordinates. The menu's position now updates with viewport changes, and its initial placement is centered below the output area. Also refactored logic to show the menu with new layers instead of adding to an existing batch.
Introduces an addLayers method to BatchPreviewManager for adding new layers to an active batch preview or showing the UI if inactive. Updates Canvas to use addLayers instead of show, and fixes a bug where new layers were only added if more than one was present.
Introduces BatchPreviewManager for reviewing and confirming multiple imported layers after auto-refresh. Adds a toggle button for mask overlay visibility in the UI and updates mask rendering logic to respect overlay visibility. Also refactors image import to return new layers and adds a utility for removing layers by ID.
Introduces a new backend route and method to fetch all images created since a given timestamp, and updates the frontend to import all new images as layers on auto-refresh. This improves workflow by allowing multiple images generated in a single execution to be imported at once, rather than only the most recent image.
Introduces an 'Auto-refresh after generation' toggle to the Canvas. When enabled, the latest image is automatically imported after a successful execution event. Also ensures event listeners are properly cleaned up when the node is removed.
Centralizes layer movement logic in CanvasLayers with a new moveLayers function, supporting both up/down and drag-and-drop reordering. Updates selection logic in Canvas to only trigger updates when selection changes, and improves event handling in CanvasLayersPanel for more responsive selection and drag operations. Removes redundant moveLayersToPosition method in favor of the new unified approach.
Introduces a new CanvasLayersPanel component for managing layers visually, including selection, renaming, reordering via drag-and-drop, and deletion. Integrates the panel into the main Canvas and CanvasView, synchronizes selection and state changes, and adds logic for duplicating layers and debounced state saving. Moves IndexedDB state saving to a Web Worker for better performance. Also sets default logger level to DEBUG for improved diagnostics.
Added documentation files for ComfyApi, ComfyApp, LitegraphService, and MaskEditor, summarizing their main functions and usage. Refactored js/Canvas.js to improve mask processing logic, using viewport pan for cropping and applying mask color only to non-transparent pixels. Also made minor formatting and logging consistency improvements throughout Canvas.js.
Refactored the Canvas class to act as a facade, providing a simplified high-level interface and delegating detailed operations to internal modules. Added Polish documentation, grouped and clarified main operations, and moved legacy/delegation methods to the end for backward compatibility. Removed unused or redundant methods from CanvasLayers.js, such as removeLayer, moveLayer, addMattedLayer, isRotationHandle, getResizeHandle, handleBlendModeSelection, and getFlattenedCanvasAsDataURL, to streamline the codebase.
Introduces the ability to open the current canvas in a mask editor, upload and retrieve mask edits, and apply them to the mask layer. Adds utility functions for mask editor state detection and control, a new 'Edit Mask' button in the UI, and methods for handling mask updates and preview refresh. Also adds a setMask method to MaskTool for precise mask placement.
Introduces a 'fit_on_add' boolean option to control whether images are fit to the canvas when added or pasted. Updates image addition and paste logic in Canvas, CanvasLayers, and CanvasView to support new placement modes ('fit', 'center', 'mouse', 'default').
The handlePaste method now accepts a pasteMode parameter to control image placement, allowing images to be pasted either at the mouse position or centered on the canvas. This improves user control when pasting images via keyboard shortcuts or UI buttons.
Introduces a brush preview overlay using a separate preview canvas in MaskTool. Mouse event handlers in CanvasInteractions and MaskTool are updated to support passing both world and view coordinates, enabling accurate brush preview rendering. The preview is shown or hidden appropriately on mouse enter/leave and while drawing.
Set logging level to DEBUG in both Python and JavaScript loggers for improved debugging. Updated MaskTool to accept an onStateChange callback and trigger it after mask state changes, allowing Canvas to respond to mask updates.
Replaces the onInteractionEnd callback with onStateChange for more consistent state change notifications. Adds a removeSelectedLayers method to Canvas for cleaner layer removal logic. Updates UI event handlers to use the new methods and callbacks, and ensures state is saved after relevant operations. Cleans up redundant updateOutput calls and streamlines output update logic.
Introduces an optional onInteractionEnd callback to the Canvas class, which is triggered at the end of user interactions. CanvasView now uses this callback to update output after interactions, improving responsiveness to user actions.
This commit applies consistent code formatting across multiple files, including spacing, indentation, and object destructuring. No functional changes were made; the update improves code readability and maintainability.
Introduces a WebSocket-based mechanism for CanvasNode to send and receive canvas image and mask data in RAM, enabling fast, diskless data transfer between frontend and backend. Adds a new WebSocketManager utility, updates CanvasIO to support RAM output mode, and modifies CanvasView to send canvas data via WebSocket before prompt execution. The backend (canvas_node.py) is updated to handle WebSocket data storage and retrieval, with improved locking and cleanup logic. This change improves workflow speed and reliability by avoiding unnecessary disk I/O and ensuring up-to-date canvas data is available during node execution.
Introduces an operation counter and threshold in ImageReferenceManager to trigger automatic garbage collection after a set number of canvas operations. Canvas now increments the operation count on save, undo, and redo, and exposes methods to set the operation threshold and retrieve stats including operation count. CanvasView displays the operation count and threshold after manual garbage collection.
