FC FaceDetection

FC FaceDetection is the bare-bones detector in ComfyUI-FaceChain: it finds a face in an image and hands you its bounding box - but doesn't crop, mask, or do anything else. Two inputs in, an image and a box out. It's the "just tell me where the face is" node.

If that sounds like a stripped-down FC FaceDetectCrop, that's because it is - this is one of the pack's older/underscore-class names, from before the current node set consolidated detection into FC FaceDetectCrop (which does detection and crop and mask and landmarks in one node). It still shows up in older workflows and on pack listings, so knowing what it does and where it plugs in is worth having even if you wouldn't start a new graph with it.

Inputs and output

• source_image (IMAGE) - the photo to search.
• face_index (INT, default 0) - which face to return when there are several, by detection order.

Outputs:

• IMAGE - the source image, passed through unchanged. (This is a quirk worth knowing: the node returns the input rather than a crop - the box is the real payload.)
• BOX - the face bounding box in original-image coordinates.

The engine is the same ModelScope face-detection pipeline (damo/cv_ddsar_face-detection_iclr23-damofd) the rest of the pack uses, downloaded on first use from modelscope.cn.

Where it fits

The BOX output is what you want. It plugs into anything that consumes a BOX - a FC CropMask-style crop node, or this pack's FC CropToOrigin for paste-back alignment. If your graph already has a detector doing the crop but you also need the raw box for another branch (say, to mask something near the face), this node supplies it without re-running the whole crop pipeline.

The honest caveats

Two things. First, the node's IMAGE output being the unchanged input trips people up - if you expected a face crop, you'll stare at a full portrait and think it's broken. The crop lives in FC FaceDetectCrop; this node only locates. Second, because it's an older class name, loading a workflow that references FC_FaceDetection may complain about a missing node after a pack update - the current mapping uses FC FaceDetectCrop instead. Both work, but the current name is the one you want for new graphs.

Install

Same pack, same drill: ComfyUI Manager (search "ComfyUI-FaceChain") or git clone https://github.com/THtianhao/ComfyUI-FaceChain into custom_nodes/, restart. The face-detection model downloads on first run, and the pack's heavy startup dependency install (onnxruntime-gpu, mmdet, mmcv, modelscope, insightface, transformers…) applies pack-wide.