shootthesound/comfyui-angelo

Angelo is a custom node for ComfyUI, the popular open-source tool for running image generation models like Stable Diffusion and FLUX. ComfyUI normally works by wiring boxes together into a graph: a model loader, a sampler, a mask editor, and so on. Angelo collapses the usual generate-then-fix loop into one node. You generate an image, then click, paint a stroke, or drag a rectangle directly on the node's built-in preview to refine just that part. The rest of the image stays bit identical. The author built it for the FLUX 2 Klein 9B distilled model with 4 step sampling and CFG of 1, but it works with any sampler-compatible model. The defaults are pre-tuned for Klein, and the repo ships an example workflow file you can drag onto the canvas to get a complete graph wired up. The node has two modes, switched by a dropdown. In Sampler Mode it behaves like a regular sampler and produces the base image from the incoming latent. In Edit Mode the click and paint controls turn on and the seed is locked to whatever produced the cached image, so refines stay coherent. There is also a Load Image button that lets you edit an existing photo: it encodes the file with the connected VAE so you do not need to wire up a separate VAEEncode node. Beyond plain click-to-refine, the node bundles several edit styles. An Area Prompt text box inside the node lets you refine a region with a different prompt than the main one, for example fixing a face while keeping the scene prompt for the rest. Smart Inpaint adds new content into a rectangle you drag using an edit model. Smart Guided Inpaint skips drawing and instead takes a location from a dropdown like "top left" along with a description. An optional Detect feature uses SAM 3 so you can type "the face" and have the silhouette masked for you. Xtra Fine, Undo, Reset, and a Persistent Mask toggle for generating variations round out the controls.