ComfyUI-WanAnimatePreprocessV2
Improved Pose & Face Detection for Wan Video 2.2 Animate based on GitHub Issue 1410 (WanVideoPreProcessor) and 10 (WanAnimatePreProcessor). (Special thanks: @kijai and @steven850)
Nodes (16)
Six conditioning maps from one node, when you can't decide what the ControlNet wants
Turn detected poses into the skeleton image your Wan-Animate sampler actually sees
Count blinks on your driving video so you know whether your animated character blinks at all
No, the dropdown isn't broken — put the .onnx files in models/detection
The node that decides whether your Wan Animate drive looks alive or wrong
Track pupil dilation without the camera-distance trap
It classifies saccades at 300°/s — the eye twitches your character animation keeps losing
Your face crops are feeding the encoder garbage — this node catches it before it does
Twelve expression dials, 3-DoF head rotation, gaze offset — all on the pose data, before the sampler
ETH-XGaze at ~2.5° mean angular error
Render eye-gaze into a ControlNet conditioning image — very experimental, very specific
Your Wan Animate skeleton is shaking — this node gives it a 3D spine
Just the skeleton, please — keypoints without the face-crop and gaze machinery
Reshape an OpenPose-18 skeleton into BODY-25, COCO-17 or MediaPipe-33 without re-detecting
A 0-to-1 quality score for your driving clip, before you burn an hour on a render
Fit the lighting of one shot and relight another — spherical harmonics, no model required
Improved Pose & Face Detection for Wan Video 2.2 Animate based on GitHub Issue #1410 (WanVideoPreProcessor) and #10 (WanAnimatePreProcessor). (Special thanks: @kijai and @steven850)