SDPose-OOD: Video to Pose Map
Upload a video to extract pose keypoints and generate a pose map. The workflow supports multiple person detection and uses an enhanced SDPose model for accurate whole-body feature extraction.
About
This ComfyUI tutorial walks you through SDPose-OOD: Video to Pose Map — a workflow that converts any input video into a clean pose map sequence. The pipeline starts with LoadVideo to decode frames, runs each frame through the SDPose-OOD pose extractor (backed by the sdpose_wholebody_fp16.safetensors checkpoint) to detect whole-body keypoints, and then assembles the pose frames into a final clip with CreateVideo and SaveVideo. The SDPose-OOD node (type ID 01b6a731-fb78-4070-9a38-c87146da9604) supports multi-person detection and outputs a rendered pose map per frame, covering body, hands, and face landmarks.
What you provide
- Reading the workflow
Runs in ComfyUI. This workflow can't run on this page. Open it in Comfy Cloud with everything loaded, or download it for your own ComfyUI.
Try in CloudQuestions & answers
Yes. The SDPose-OOD node supports multi-person detection and will render pose maps for all detected subjects per frame. For crowded scenes, consider raising the confidence threshold to reduce false positives.
Use sdpose_wholebody_fp16.safetensors. Place it in ComfyUI/models/checkpoints/. Then select it in the SDPose-OOD node before running the workflow.
Pass the frame rate from LoadVideo into CreateVideo, and keep the output resolution consistent with the input. This preserves one-to-one timing and sizing for downstream use.
Try increasing the input resolution for clearer details, raise the detection confidence to reduce noise, and ensure good lighting and contrast in the source. For performance-constrained GPUs, balance resolution and frame skip to maintain stability.