OpenPose Part Stitch
Paste your detailed crop back in — and make the seam vanish
- stitch
- image
- blend_mask
- image
You cropped a hand with OpenPose Part Crop, ran it through a detailer at 1024px, and now you need to put the freshly-detailed pixels back where they came from. That's the entire job of this node, and it's the third corner of the pack's pose-driven detailer loop: crop, process, stitch.
How it works
Everything the Stitch node needs to place the crop correctly is carried in the stitch handle that OpenPose Part Crop produced - the original image, the native region rectangle per frame, the feather setting, and the up/downscale methods. Stitch takes that handle plus your processed image, resizes the processed crop back to its original native box (using the recorded scale method), and blends it in with a feathered rectangle mask. The feather is the same mask_feather value the Crop node used, so the two halves agree on the blend by construction - you never have to re-enter a pixel coordinate or a feather amount.
Inputs and outputs
stitch- the handle fromOpenPose Part Crop. Don't cross wires between different crops; each stitch handle belongs to one crop.image- the detailed/processed crop to paste back. Same size as the crop you got out of the Crop node, or close to it - the node resizes it either way.blend_mask(optional) - an override mask the same size as the processed crop. If you don't want the default feathered rectangle - say your detailer output has an edge you want to preserve, or you want to blend a non-rectangular region - supply your own here.
Output is a single image: the original frame with the processed region blended in. Batch-aware, so it handles multi-frame sequences the same way the crop side does.
The workflow
OpenPose Part Crop → your detail model (an inpaint pass, a hi-res fix, a face/hand ControlNet at real resolution) → OpenPose Part Stitch. The pattern is the same crop-edit-stitch shape that became the default way to fix a region in ComfyUI - crop at native size gives a small region a huge generation budget, and stitching leaves the rest of the frame untouched, not even through a VAE round trip. Compared to an all-in-one auto-detailer like Impact Pack's FaceDetailer, this is the manual, pose-targeted version - more setup, but you control exactly which region gets detailed.
Install
Part of Mickmumpitz-Nodes (MIT, deps: numpy, Pillow, opencv-python - no models). ComfyUI Manager, search "Mickmumpitz", or:
cd ComfyUI/custom_nodes && git clone https://github.com/mickmumpitz/ComfyUI-Mickmumpitz-Nodes.git
Restart, find it as "OpenPose Part Stitch" under Mickmumpitz/OpenPosePartCropper. It's the last 30 seconds of the loop, but it's the part that decides whether the fix looks native or looks pasted - and it makes "pasted" the hard option.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| stitch | OPP_STITCH | — | |
| image | IMAGE | The detailed/processed crop to paste back. | |
| blend_maskopt | MASK | Optional override mask (same size as the processed crop). |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |