ComfyUI-segment-anything-2

Why we included this project

For ComfyUI users who need to cut objects out of images or video frames, this pack wraps Meta's Segment Anything 2 into a set of nodes. You can run promptable segmentation directly on images and video, then feed the resulting masks into downstream nodes for inpainting or compositing. Models download automatically from Hugging Face into the standard ComfyUI model folder, so setup stays simple for anyone used to the ecosystem. The project is upfront about being work in progress: mask postprocessing is currently disabled and the coordinate selector still needs refinement, so it's best for tinkerers who are comfortable assembling their own node chains rather than expecting a finished one-click tool. For segmentation-heavy work inside ComfyUI, it saves you from wiring SAM2 up by hand.