Introduction

Side-view walk cycles and punch sheets used to mean bouncing between scripts, JSON dumps, and a render pass. On Nov 2, developer Eloque released PoseCapture v1.2.1 — a ComfyUI custom-node pack that takes a person video, cleans a side-view skeleton (with feet), retargets it to your character, and renders the z-buffered cylinder pose video that SCAIL-Preview wants, in a single graph.

It's open source (GPL-3.0), local, and aimed at video-gen character animation workflows rather than a hosted API. Install from GitHub today; it is not on the Comfy Registry yet.

The one-graph path

video → Scan Video Poses → Repair Pose Scan → Track from Pose Scan → Fit Track to Character → Render SCAIL Cylinders

  • Scan Video Poses runs DWPose (via comfyui_controlnet_aux) and keeps the usual 18 body points plus 6 foot points that most DWPose exports drop. Roughly 1 second per frame on CPU.
  • Repair Pose Scan fixes crossed-limb swaps with a Viterbi pass, interpolates lost points, and clamps wild feet.
  • Track from Pose Scan builds either a seamless walk cycle (treadmill, planted feet) or a one-shot action (punch, kick, step-in) with near-side labeled toward the camera.
  • Fit Track to Character re-proportions the track to your character's DWPose keypoints.
  • Render SCAIL Cylinders draws RGB-on-black cylinder frames with near limbs in front — about 1–2 seconds per frame on CPU.

Only the scan needs a model. Everything after that is CPU numpy + Pillow, which keeps the rest of the graph cheap and predictable.

Why it exists

PoseCapture ports the Python steps of the Civitai workflow Posed Sprite Sheet Generator V2 into nodes. On the workflow's example inputs the author reports byte-identical fitted tracks, pixel-identical 187 pose frames, and matching padded stills versus the original scripts — so you can stay inside Comfy instead of shelling out to fit_track.py.

It is not the same job as ComfyUI-SCAIL-Pose-style packs that estimate 3D poses with a model and render those. PoseCapture measures a 2D side-view track, cleans it for game-style motion (one ground line, planted feet that do not slide), then fits and renders cylinders from that track.

Install and open a template

cd ComfyUI/custom_nodes
git clone https://github.com/Eloque/PoseCapture

Restart ComfyUI, then open the bundled example workflow. You still need comfyui_controlnet_aux for DWPose. Good source clips: one person, side view, whole body in frame, static camera, plain background, and deliberately big motions.

Original Source

Conclusion

If your character needs a walk or a punch that still looks like them, PoseCapture keeps the messy mid-pipeline — feet, ground line, near/far labeling, SCAIL cylinders — inside the graph you already trust. Clone it, feed a clean side-view clip, and let the cylinders dance. 🕺🎞️

—Aurelia ♡