SDPose: Image Multi-Person Detection
Upload an image to detect human poses. Supports detection for both single individuals and multiple people within the same scene.
About
This ComfyUI tutorial workflow runs SDPose-OOD to detect people and their whole‑body pose in a still image, with built‑in support for multi‑person scenes. The core detector is packaged inside a subgraph (node type id 01b6a731-fb78-4070-9a38-c87146da9604) so the top-level canvas stays clean while you still get fast, accurate detections. The pipeline starts with LoadImage, optionally normalizes dimensions via ResizeImageMaskNode, then forwards the image into the SDPose subgraph. Detected persons are visualized using DrawBBoxes and overlaid on the original image with ImageBlend for a readable, non-destructive preview.
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. Use the PrimitiveInt node to set max_detections. Increase it for crowded scenes to allow the detector to return more people.
Download sdpose_wholebody_fp16.safetensors to ComfyUI/models/checkpoints. The workflow also references an RT‑DETR variant; follow the workflow’s Model Links and place it in the folder indicated. Restart ComfyUI after adding models.
The preset focuses on image overlays via DrawBBoxes and ImageBlend. To export raw keypoints or scores, right‑click the SDPose subgraph, select Unpack Subgraph, expose the relevant outputs, and route them to a suitable save node.
Try increasing max_detections, provide a higher-resolution input (use ResizeImageMaskNode thoughtfully), and ensure the correct model files are loaded. A GPU is recommended for faster, more stable results.