Draw SDPose Keypoints (V2)

Pose detection gives you numbers - keypoint coordinates, confidence scores, people lists. Video models like WanAnimate don't want numbers, they want a picture of the skeleton. Draw SDPose Keypoints (V2) is the pack's renderer: it takes a POSE_KEYPOINT sequence and draws the classic whole-body stick figure - body, hands, face, and optionally feet - frame by frame, out as an IMAGE batch you can feed directly into WanAnimate's pose_video input or save as the pose video itself.

It's the V2 because it layers in the stuff that makes a stick figure readable to a model instead of just technically correct. The headline features:

• Layered drawing with occlusion: it draws front-side and back-side bones in the right order so limbs cross believably instead of z-fighting.
• Yaw-aware thickness (enable_yaw_thickness): when you feed in a yaw_angles sequence (from the pack's Estimate Yaw nodes), bones get thinner as the subject turns sideways and thicker facing the camera - matching how the model perceives a 3D body in 2D. Without yaw input it just draws at a fixed stick_width.
• Two foot modes (foot_mode): dots (three colored dots at the ankle/toe points) or line (an ankle-to-toe bone line). The line mode was added for clearer foot orientation in dance/motion work.

What you set

• keypoints - the pose sequence (required).
• draw_body / draw_hands / draw_face / draw_feet - which parts to draw. Feet default off; turn them on for dance.
• stick_width (4), face_point_size (3), hand_point_size (4) - visual weight dials.
• score_threshold (0.3) - drop keypoints below this confidence; raise it to clean up jittery detections.
• max_frames (2500) - safety cap; longer inputs get truncated with a warning.
• mouth_mode - draw_all, no_draw, or inner_lip_only. The "inner lip" option hides the outer lip line, which some models read more cleanly.
• color_scheme - v4_custom (default), standard (classic OpenPose rainbow), or monochrome.
• yaw_angles (optional) - drives thickness and occlusion ordering.
• hand_scale (optional) - scales hand bones outward from the wrist (for exaggerated hand clarity) without shifting the wrist.

Output is a single IMAGE batch - black background, colored bones, ready for WanAnimateToVideoCustom's pose_video.

Install

Part of user2318/ComfyUI-CustomNodeKit. ComfyUI Manager: search "CustomNodeKit". Or:

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

then restart. It draws with numpy - no extra models. The POSE_KEYPOINT input comes from the SDPose ecosystem (the official sdpose-ood pack), which does need its own pose models.

The gotcha

The canvas size comes from the keypoints' canvas_height/canvas_width, and the whole sequence is drawn on that one canvas. If your segments have inconsistent canvases (see the resize/concat notes in this pack), the output silently inherits the first frame's size and later frames can be clipped. And don't expect this to fix bad input: score_threshold only hides low-confidence keypoints, it doesn't invent missing ones - a subject with a dropped arm will render a skeleton with no arm, and WanAnimate will happily animate it that way. If you get clean poses in the preview but janky animation, the answer is usually upstream (better detection, fix_empty_frames, a higher threshold), not more draw settings.