The centre/ driving video was a post made by me n ü, recorded near where I was living in Canada.

The other two videos show the outputs of a comfyUI workflow that extracted the poses of the source video, with open pose, and used that as the driving material for the final output videos.

The resulting videos were generated using Wan2.1 with the open pose control net input.

It was interesting to see how many errors there were in the pose extraction (Limbs appearing and disappearing) to then see how flawless the final poses were, despite missing information.

From a single driving video, unlimited final videos could be created and synchronised with BPM.