A ComfyUI custom node toolbox focused on multi-image references, SAM3/SAM3.1 person cutouts, SCAIL-2 mask workflows, image batch utilities, boolean routing, and group-based workflow switching.

Current version: v2.4. The latest version of ManualRefCollage is v2.2.2.

中文说明

Search for ComfyUI-FeiHou-Toolbox in ComfyUI Manager, install it, then restart ComfyUI.

Open your ComfyUI custom_nodes directory and run:

git clone https://github.com/FX-FeiHou/ComfyUI-FeiHou-Toolbox.git

After installing or updating, restart ComfyUI and hard-refresh the browser page.

Node display titles follow the current ComfyUI language setting. This English README shows the English titles.

An enhanced version of ComfyUI's original SCAIL-2 colored mask node, built for mask routing, separation, and batch handling in multi-reference workflows.

prefix_mask_mode supports:

• Multi Image Single Color: each output batch image is single-color, matching the current reference_image_mask foreground color.
• Multi Image Multi Color: each output batch image uses the node's object color palette rule.
• Single Image Multi Color: output batch images use batch-order colors: blue, red, green, magenta, cyan, then loop after 5 images.

Automatically uses SAM3/SAM3.1 to cut people out from multiple reference images and compose them into a single multi-person reference collage.

A manual collage node that loads SAM3/SAM3.1 cutouts onto an editable canvas, allowing users to adjust position and scale before outputting the final composed image.

A revised switch node for toggling between two workflow paths while reducing issues caused by inactive branches.

An input-only boolean node that inverts true to false and false to true. It is compatible with PrimitiveBoolean output chains.

A copy-style variant of KJNodes ImageBatchMulti for creating a batch from multiple image inputs. It keeps the dynamic input-count workflow, but skips missing or bypassed image inputs instead of adding black placeholder frames. If no image input is available, it returns no image output.

A group bypass and switch node that binds two ComfyUI Groups, switches which group is active, bypasses the inactive group, and outputs the corresponding input.

• Added ImageBatchMultiV2 (Image Batch Multi V2 / 图像组合批次（多重）V2), based on KJNodes ImageBatchMulti.
• Kept the original dynamic inputcount and Update inputs workflow.
• Changed missing-input handling: unconnected or bypassed image inputs are skipped instead of being replaced by black frames.
• If no image inputs are available, the node returns no image output.

• Added InvertBoolean (Invert Boolean / 反转布尔值), a single-input single-output boolean inverter.
• The node is compatible with PrimitiveBoolean output chains and flips true to false, false to true.

• Changed ManualRefCollage size handling so the manual canvas reads and applies current width / height inputs only when 应用尺寸 is clicked.
• ManualRefCollage no longer auto-applies linked width/height during node creation, connection changes, or 载入图片, avoiding early reads before KJNodes Set / Get variable chains are ready.
• Added compatibility for KJNodes-style Set_variable / Get_variable width and height routing; clicking 应用尺寸 now attempts to resolve the matching Set node and read the value from its input chain.

• Fixed ManualRefCollage not correctly reading externally connected width / height inputs in both the frontend manual canvas and backend execution.
• Manual collage now prioritizes linked width/height inputs and falls back to the node's own width/height widgets only when linked values cannot be resolved.

• Updated Create SCAIL-2 Colored Mask V2 prefix_mask_mode behavior.
• Multi Image Single Color renders every output batch image as a single color matching the current reference_image_mask foreground color.
• Added Multi Image Multi Color for rendering every output batch image with the node's object color palette rule.
• Single Image Multi Color renders output batch images with batch-order colors: blue, red, green, magenta, cyan, then loops after 5 images.

• Fixed ManualRefCollage preview loading images from ignored/bypassed inputs.
• Fixed ignored/bypassed image inputs still being included during manual collage execution.
• Fixed externally connected width / height being overridden by stale layout_json dimensions.

• Added FastGroupsBypassSwitch (多框忽略并切换).
• Supports enabling, bypassing, and output switching between two ComfyUI Groups.
• UI follows the rgthree Fast Groups Bypasser style for large grouped workflows.

• Finalized ManualRefCollage (多参图像手动拼接).
• Supports SAM3/SAM3.1 cutouts and manual layout for up to 5 reference images.
• Supports using the first video frame as a composition reference while keeping the final output on a black or white background.

• Initial release.
• Added Create SCAIL-2 Colored Mask V2.
• Added AutoRefCollage.