ComfyUI-Angelo Brings Click To Fix Editing To AI Image Generation

By vramkickedin | | 2 min read
A featureless white mannequin bust head is turned to the side with fingertip gently touching cheek.

ComfyUI-Angelo merges an image sampler and inpaint refiner into a single custom node for ComfyUI. The tool lets users click or paint directly on a generated image to fix specific regions, while the rest stays pixel-perfect. It replaces the usual chain of KSampler, mask editor, and post-processing with one streamlined workflow.

Shootthesound who also made ComfyUI-Wan-I2V-Control and ComfyUI-ReferenceLatentPlus, developed the node to eliminate the slow process of saving images, opening a mask editor, and re-queueing just to fix a small detail. He designed Angelo to support FLUX 2 Klein 9B and Qwen-Image-Edit as first-class edit models, with compatibility for any sampler-compatible model. By collapsing multiple steps into instant point-and-click edits, he brings a photo-editing feel to AI generation inside ComfyUI.

One node to generate and edit

Key Features
  • Click any region to refine that spot immediately.
  • Paint freeform strokes for custom mask shapes.
  • Area Prompt for region-specific text guidance.
  • Xtra-Fine upscales small details for better quality.
  • Smart Inpaint adds new objects via rectangle drag.
  • Smart Guided Inpaint places content by location phrase.
  • Optional SAM 3 detect for text-based segmentation.
  • Undo, Redo, Re-roll, and Persistent Mask controls.

This tool is ideal for photographers, designers, and AI artists who need to fix local flaws without disrupting their creative flow. It replaces what once required multiple nodes and a separate mask editor with a single, fast interface that runs entirely offline on local GPUs. Privacy-conscious professionals can use it confidently with their own models like FLUX 2 Klein or SDXL.

Developer notes and known limits

All refinements live only in ComfyUI’s current session, so restarting the software clears the cache. The Xtra-Fine upscale feature adds about 1.5–2 seconds per click on a 5090 GPU due to VAE processing. Peter Neill released the project under the MIT license and encourages bug reports and feature ideas via GitHub.

“I'm a photographer who kept hitting the same wall in ComfyUI: generate an image, then to fix one thing I'd save it, open a Mask Editor or Photoshop, and fix. It works, but it's not smooth.” — Source: Reddit

Project Links