[Inference.Core] Save Pose Keypoints

This pack's pose detectors - DWPose, OpenPose, AnimalPose - all give you two things: a rendered skeleton IMAGE and a POSE_KEYPOINT output holding the actual coordinate data behind it. Most workflows only ever touch the image, wire it to a ControlNet, and move on. This node is for the other case: when you want the raw keypoint data itself, saved as a JSON file, rather than just the picture. That's useful for archiving a pose alongside a generation for later reuse without re-running detection, feeding the data into an external pose-editing tool, or building a small library of poses you can pull from a script instead of a reference image.

How it works

It's an output/terminal node - it doesn't return anything to the graph, it writes a file to disk and stops. It takes whatever POSE_KEYPOINT structure a detector node produced and serializes it directly to JSON, preserving whatever detail level that detector captured (body-only, or body+hands+face, depending on how the upstream node was configured).

The inputs and outputs that matter

• pose_kps - required, the POSE_KEYPOINT output from any of this pack's pose detector nodes (DWPose Estimator, OpenPose Pose, AnimalPose Estimator).
• filename_prefix (default PoseKeypoint) - the prefix used for the saved JSON filename.

No outputs - this is a terminal node (is_output_node: true), meaning it does its job as a side effect (writing the file) rather than passing anything further down your graph.

How to install it

ComfyUI Manager: search ComfyUI-Inference-Core-Nodes, install, restart. By hand:

cd ComfyUI/custom_nodes
git clone https://github.com/LykosAI/ComfyUI-Inference-Core-Nodes

then run install.py, or pip install -e . - no acceleration extras needed for this one specifically, since it's a pure file-writer with no model inference. Restart ComfyUI.

Common issues & troubleshooting

The most common surprise: nothing visibly happens in the graph. Because this is a terminal node with no image or data output, if you expect to see a preview or chain something after it, you won't - check your ComfyUI output directory for the saved JSON file instead of looking for a node preview.

Make sure you're actually feeding it a POSE_KEYPOINT output and not the IMAGE output from the same detector node - they come out of the same pose nodes side by side, and it's an easy wire to grab by mistake since both represent "the pose" conceptually. Only POSE_KEYPOINT will connect here; ComfyUI's type checking will refuse the IMAGE wire outright, so a failed connection here is usually just the wrong output socket, not a broken install.