ComfyUI-Impact-Pack

Maintained by ltdrdata, who also created ComfyUI-Manager, this extension adds ComfyUI nodes built around a detect, crop, regenerate, and paste-back loop. Detector nodes find faces, people, or arbitrary regions with SAM or SAM2 segmentation, ONNX models, or CLIPSeg, and package each hit as a SEGS object holding a mask and crop box. Detailer nodes such as FaceDetailer then re-sample every region at higher resolution with your checkpoint and composite it back, repairing the mangled small faces typical of full-body diffusion renders. Other groups cover iterative multi-step upscaling, regional sampling with a separate prompt per mask, wildcard and dynamic prompt files in txt or yaml, mask arithmetic, per-frame video detection, switches, and pipe objects bundling model, CLIP, and VAE links. YOLO detection via Ultralytics sits in the separate Impact Subpack. Install through ComfyUI-Manager or clone into custom_nodes and add the requirements; current releases expect a recent ComfyUI build. Licensed GPL-3.0 like ComfyUI. Best for diffusion users who want automated face and detail inpainting without hand-masking every image.