ComfyUI Node
Scan Video Poses (DWPose + feet)
Runs DWPose whole-body on every frame of a video and keeps the 18 body points AND the six foot points (heel, big toe, small toe of each foot), which the usual DWPose keypoints drop: the feet give the ground contact. Connect the frames of a Load Video node to images (or, for a long video, leave images unconnected and give a video file path: it is then read frame by frame without holding the whole clip in memory). crop = x,y,w,h of the person's part of the frame (empty = the whole frame). About 1 s per frame on a CPU. Needs comfyui_controlnet_aux.
Scan Video Poses (DWPose + feet)
- images
- scan
- overlay
- info
◄crop►
◄start0►
◄frames0►
◄every1►
◄bbox_detectoryolox_l.onnx►
◄pose_estimatordw-ll_ucoco_384_bs5.torchscript.pt►
◄overlay_height720►
◄video►
CategoryPosed Sprite Sheet
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| crop | STRING | x,y,w,h in pixels of the frames; empty = whole frame | |
| start | INT | 00–100000 | first frame |
| frames | INT | 00–100000 | frames to scan, 0 = to the end |
| every | INT | 11–100 | take every n-th frame |
| bbox_detector | COMBO | yolox_l.onnx | 2 options: yolox_l.onnx, yolox_l.torchscript.pt |
| pose_estimator | COMBO | dw-ll_ucoco_384_bs5.torchscript.pt | 2 options: dw-ll_ucoco_384_bs5.torchscript.pt, dw-ll_ucoco_384.onnx |
| overlay_height | INT | 7200–4096 | height of the overlay frames (0 = scan size) |
| imagesopt | IMAGE | the video frames (from Load Video) | |
| videoopt | STRING | used only when images is not connected: a video file, relative to ComfyUI/input or absolute |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| scan | PSS_POSE_SCAN | — |
| overlay | IMAGE | — |
| info | STRING | — |