OpenPoseKeyPose

OpenPoseKeyPose is the CUP pack's loader for keypose control images - and it's the one node in this family where the difference in the folder it reads actually changes your workflow. The rest of the OpenPose siblings fetch a rendered control image for every animation frame. Keypose is where you don't want that: keyframes are the handful of poses that define a motion, and everything between them is interpolation. Animators think in keyframes; this node lets the pipeline feed ControlNet exactly those, so the model poses characters at the important beats instead of chasing every intermediate frame.

It's an OpenPoseBase subclass with posetype = "keypose", so it reads <image-dir>/keypose/ImageNNNN.png. Like every node in this pack, it does zero pose detection - the Blender addon (ComfyUI-BlenderAI-node) renders the keypose images, and this node loads them into your graph.

How it works

• image (STRING) - directory of rendered control images, filled by the Blender addon.
• frame (INT, default 0) - which keypose image to load.

Outputs: IMAGE (the keypose drawing → Apply ControlNet with a pose model) and MASK (from the image's alpha, optional). The frame input still indexes into the folder, but you'll typically set it to the keyframes you care about rather than stepping through a full timeline - that's the whole point of having a separate keypose type at all.

Install and gotchas

Ships with ComfyUI-CUP. Manager search ComfyUI-CUP, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AIGODLIKE/ComfyUI-CUP.git

Restart after; no models or pip packages to worry about.

Same family rules: loader, not detector; dead in a plain web workflow without the Blender addon feeding it; and a missing frame returns a silent 64×64 black placeholder that will quietly corrupt a ControlNet pass. If your keyposed poses stop landing, verify keypose/ImageNNNN.png exists first.