A convulted workflow that takes an input video 512x512 x32 frames

Creates a Canny, Openpose, and Depth control nets.

Pushes it through animatediff pipeline

and finally a face detailer to unmuck faces.

removed batch prompt travel

working on replacing the face detailer with a consistent face mask detect, and inpaint from a single frame extreme close up face shot generated first.

From the Load Video Path, I suggest setting your select every nth to a divisor value that will return 32 frames from the source video

You may can check how many frames are being generated by checking the Pose Control Preview Images