SCAIL CMU Sequence to Pose

The CMU nodes store their motion data in their own internal format, and SCAILCMUMotion hands that to you as a SCAIL_POSE_SEQUENCE. Sometimes that's all you need - chain it straight into the renderer. But the format is meant for playback, and when you want the motion as a proper frame-by-frame SCAIL pose set with correct limb structure and colors for a known canvas size, you run it through SCAILCMUSequenceToSCAILPose. It's a conversion node: in one end, out the other, same type on both sides, but the payload is rebuilt for rendering.

Two inputs, both required:

• pose_sequence - the SCAIL_POSE_SEQUENCE from SCAILCMUMotion (or a chain of them).
• image_width and image_height - the canvas dimensions (512×768 defaults) the sequence is being prepared for. They don't resize anything; they record the projection context so the renderer and downstream pose consumers know what space the sequence was authored for.

The node stacks all the frames into a single tensor, rebuilds the OpenPose 18-limb structure per character (with the pack's standard per-bone colors), and tracks the character count and frame count. Output is a single pose_sequence of type SCAIL_POSE_SEQUENCE - and the practical point of feeding it through is that the output carries clean per-frame joints plus limb_seq and colors that SCAILPoseRenderer and any pose-analysis node can consume without guessing at the internal motion format.

Honestly, for the basic Walk → Run → Jump → Render graph from the README, you don't strictly need this node - SCAILCMUMotion's sequence output already renders. It becomes relevant when you're doing something less trivial: inspecting the joints directly, blending a CMU motion with other pose sources, or standardizing the payload before it hits a custom pose consumer that expects the canonical structure. It's the kind of node you reach for when the render starts complaining about an unexpected structure, or when you want one consistent pose format across the whole graph.

Installation is the shared pack story: clone ckinpdx/ComfyUI-SCAIL-AudioReactive into ComfyUI/custom_nodes, pip install -r requirements.txt, restart, or use ComfyUI Manager. No downloads of its own - it only does its work on data that's already in memory, which is why it's effectively free to run even on long chains.