ARC Composite

ARC Composite v0.2.0

A lightweight, model-agnostic ComfyUI compositor for blending generated image edits back onto a source image while preserving untouched regions.

ARC Composite is designed as a post-processing step for image-editing workflows. The edited image can come from an API model such as Nano Banana / Gemini or GPT Image, a local image-edit model, or any other workflow that outputs a standard ComfyUI IMAGE.

The node does not depend on the model that produced the edit. It works from the source image, generated image, and edit mask, then builds a composite mask, feathers the transition, and optionally performs simple seam color matching.

Typical workflow

Source Image + Edit Mask
          |
          v
Image-edit model / API
(Nano Banana, GPT Image, local model, etc.)
          |
          v
    ARC Composite
          |
          v
      Final Image

Inputs

• generated_image — edited/generated image returned by an API or local model
• source_image — original source/canvas image to preserve outside the composite region
• edit_mask — authored edit mask
• mask_mode — ARC Mask / ARC Mask + Detected Changes / Detected Changes
• change_sensitivity — threshold used when detecting differences between source and generated images
• mask_grow_px — expands the working composite mask
• edge_feather_px — softens the transition between source and generated regions
• seam_match — applies lightweight color-shift matching near the composite seam
• debug — outputs a diagnostic comparison view

Outputs

• image — final composite
• composite_mask — mask used for the composite
• report — text summary of the composite operation
• debug_view — optional source/generated/mask/result diagnostic view

Notes

• The generated image is automatically resized to the source image dimensions when needed.
• Best results come from generated edits that remain spatially aligned with the source image.
• Difference detection can be used by itself or combined with the authored edit mask.
• Pure PyTorch implementation with no OpenCV, SciPy, or scikit-image runtime dependency.

Compatibility

ARC Composite operates on standard ComfyUI IMAGE and MASK types, so it is not tied to a specific image-generation model or API.

Examples include:

• Nano Banana / Gemini image editing
• GPT Image editing
• Local image-edit models
• Other API or custom workflows that return a ComfyUI image

License

GPL-3.0

Author

Developed by DHan315.