Zenkai Video Pose

Want to animate someone dancing, fighting, or just moving, without hand-drawing a single keyframe? The standard trick is pose conditioning: feed the video model a sequence of pose maps and let it follow them. The hard part is having that sequence. Zenkai Video Pose exists to make it boring: it loads a folder of pose images, turns them into a frame batch, and hands the batch to your video workflow.

It's the video sibling of Zenkai Pose Map, and it's part of the Zenkai family inside DJZ-Nodes. Where the single-image version picks one pose, this one loads a whole walk cycle.

How it works

The node scans the subfolders of the pack's posemaps/ directory, sorts the image files in your chosen folder alphabetically, then reads them as an RGB float tensor batch. The one clever bit: it cycles. If you ask for more frames than the folder holds, it wraps around with modulo - so a four-image folder and max_frames = 30 gives you the four poses repeated until you hit thirty frames.

That looping matters more than it looks. A real walk cycle is a short loop of poses repeated for the whole clip, so the modulo behavior is exactly what you want for most animation passes.

The inputs that matter

Just two:

• pose_folder - dropdown of subfolders under posemaps/, same as its single-image sibling. Drop your own folders in and restart to see them.
• max_frames - how many frames the output batch holds, default 30, up to 1000. Bigger than your folder's image count? It loops.

Output

A single IMAGE batch of shape [max_frames, H, W, 3] - the whole sequence in one tensor. Feed it into a ControlNet Apply (OpenPose/DWPose-style control model) that's patched into your video sampler, and the model gets a pose skeleton on every frame. There's no seed input here, so the sequence itself is fully deterministic: same folder, same max_frames, same output, every run.

Install

It's one of 100+ nodes in DJZ-Nodes, so install the pack once:

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/DJZ-Nodes
cd DJZ-Nodes
pip install -r requirements.txt

Restart ComfyUI afterwards, or just search "DJZ-Nodes" in ComfyUI Manager. The node's own code is just PIL/numpy/torch, but the pack's requirements pull in opencv, librosa and other heavyweights for its video/audio/3D nodes, so budget some install time.

Gotchas

• Filenames control order. Sorting is alphabetical, so pose_01.png, pose_02.png ... pose_10.png puts pose_10 right after pose_01 in a way you didn't intend. Zero-pad your names.
• It doesn't detect poses. Feed it actual pose maps (skeleton renders or OpenPose outputs). It's a loader, not a detector.
• Not a batch tool. One folder per run. If you want to sweep several folders, you're chaining or swapping the dropdown yourself.

If you're already running pose-driven image-to-video and just want a reliable way to feed it a walking sequence, this is the node. If you need a random pose per frame instead of a looping sequence, that's Zenkai Pose Map's job.