Nodes/ComfyUI-WanAnimatePreprocessV2/Wan Pose Detect — YOLO + ViTPose (V2)
ComfyUI Node

Wan Pose Detect — YOLO + ViTPose (V2)

Standalone YOLO + ViTPose detection. Takes an IMAGE batch and a POSEMODEL bundle (from OnnxDetectionModelLoaderV2) and emits a POSEDATA bundle compatible with the V2 editor and downstream conditioning nodes. No face-cropping / gaze pipeline — keypoints only.

By Code2Collapse·Created 7 months ago·Updated 12 days ago· 17
Wan Pose Detect — YOLO + ViTPose (V2)
  • images
  • model
  • pose_data
  • info
detection_threshold0.30
pose_threshold0.30
use_clahetrue
use_blur_for_posefalse
blur_radius2
blur_sigma1.5
rescale1.25
fallback_to_full_frametrue
CategoryWanAnimatePreprocessV2/extras

Inputs (10)

NameTypeDefaultDescription
imagesIMAGERGB image stack to detect poses on. Float [0,1], shape (B,H,W,3).
modelPOSEMODELViTPose+YOLO bundle from OnnxDetectionModelLoaderV2.
detection_thresholdoptFLOAT0.300–1YOLO person-detection confidence threshold.
pose_thresholdoptFLOAT0.300–1Per-keypoint confidence threshold; below this the keypoint's confidence is forced to 0.
use_claheoptBOOLEANtrueCLAHE contrast enhancement on the 256x192 pose crop. Matches the main preprocessor default.
use_blur_for_poseoptBOOLEANfalseApply a Gaussian blur to the images before YOLO+ViTPose (anti-aliases noisy frames).
blur_radiusoptINT20–32
blur_sigmaoptFLOAT1.50–8
rescaleoptFLOAT1.251–2Bbox padding factor before crop for ViTPose. 1.25 = match the main preprocessor.
fallback_to_full_frameoptBOOLEANtrueIf YOLO finds no person in a frame, run ViTPose on the entire frame instead of skipping it.

Outputs (2)

NameTypeDescription
pose_dataPOSEDATA
infoSTRING