Resample SDPose Keypoints

Load SDPose JSON resamples on the way in, but only when the file carries an fps header. Resample SDPose Keypoints is the version that works on any pose stream, anywhere in the graph, with no JSON involved. You load your keypoints at native capture speed, decide the driving video for WanAnimate runs at a different frame rate, and this node resamples the keypoint sequence to match - downsample to drop frames, upsample to synthesize them.

Under the hood it's a simple ratio: input_fps / output_fps greater than 1 means downsampling (frame dropping), less than 1 means upsampling. For upsampling you choose the interpolation method - interpolate linearly blends between neighboring keypoint positions for smooth motion, duplicate copies the nearest frame. The latter is cheaper and avoids the slightly rubbery in-between poses linear interpolation can produce on fast movements; for slow, stable motion the interpolated result is usually nicer.

Inputs and outputs

• pose_keypoints - the POSE_KEYPOINT stream (required).
• input_fps - the sequence's original frame rate. 0 means unknown, in which case nothing happens (it passes through untouched).
• output_fps - the target. 0 also means no processing. Both need to be positive for work to happen.
• interp_method - interpolate or duplicate, only for the upsampling direction.
• fix_empty_frames (optional, off by default) - same repair logic as the Load node: detects frames with empty people lists and fills them from valid neighbors. Handy here because resampling a bad frame can otherwise multiply the damage.

Outputs are the resampled pose_keypoints and an fps float (your output_fps when resampling happened, 0 otherwise). Chain it after Load SDPose JSON or any other pose source.

When to reach for it

The classic scenario: your source video was 30 or 60fps, your WanAnimate generation runs at 16fps, and you want the pose drive rate to match so the animation timing doesn't drift. It's also useful in the pack's 360° preprocessing where you assemble reference-angle pose batches at a uniform frame count. It's a dumb, honest utility - no models, no magic, just clean resampling on data you already paid to detect.

Install

Part of ComfyUI-CustomNodeKit. Via Manager (search "ComfyUI-CustomNodeKit") or:

cd ComfyUI/custom_nodes
git clone https://github.com/user2318/ComfyUI-CustomNodeKit.git
cd ComfyUI-CustomNodeKit
pip install -r requirements.txt

Restart ComfyUI after installing. Nothing extra to download for this one.