FC CropAndPaste

FC CropAndPaste is the pack's "repose the person" node. Where most of ComfyUI-FaceChain is about faces - cropping them, swapping them, pasting them back - this one is about the whole person. It takes a photo of a human and a template image, and it warps the human into the template's geometry: same person, standing where the template stood, matching its scale and pose enough to composite.

It's the fiddly middle sibling in this pack, and honestly the least polished of the crop nodes. But it's the missing piece if you want to move a person between scenes rather than just swap a face.

What it does

Two required inputs: template_image (the target - the scene and pose you want to fit into) and human_image (the person you want to paste in). It returns three outputs:

• crop_image - the human, warped onto the template's geometry
• mask - the paste mask (where the human now is)
• invert_mask - its inverse (everything else)

The mechanism is the familiar facechain chain: detect faces in both images, segment the human out of the source, then warp using a five-point facial landmark similarity transform so the person's face aligns with the template's face. That's why it tends to work on upright, face-visible portraits and falls apart on profiles, heavy perspective, or half-hidden faces - the alignment anchor is the face, not the body.

Where it fits

Use it when you have a template image (say, a studio pose) and a separate photo of a person, and you want the person composited onto the template before further generation. The mask output is the useful one for downstream steps - feed it into an inpainting setup or a paste node so the composited person doesn't bleed into the background. The invert_mask covers the opposite side of the same cut.

In practice, expect to do cleanup after this node. The warp is a geometric approximation; hair edges and clothing boundaries usually need a mask operation (this pack's FC MaskOP is right there for eroding/blurring the seam) or a quick inpaint pass.

Practical notes

• Both images should be reasonably aligned in content type - don't try to warp a headshot onto a full-body template and expect magic.
• The node calls the same ModelScope detection and segmentation models as the rest of the pack, so the first run triggers downloads from modelscope.cn (face detection + human parsing).
• It's a facechain/crop category node; install via ComfyUI Manager (search "ComfyUI-FaceChain") or clone into custom_nodes/, then restart - and budget for the pack's heavy startup dependency install (onnxruntime-gpu, mmdet, mmcv, modelscope, insightface…).

This is the node to reach for when "face swap" isn't enough and you need the whole person in a new place. Set your expectations on seam quality, keep a mask-op node handy, and you'll get decent results for a zero-training ComfyUI graph.