The mane application is making some 3D models dance by capturing motion from a video.  Obviously, openpose only captures 2D.  The rest would have to fall back on manual entry or re-enacting the source video with 2 cameras & a lion.  There might be a way to match time stamps in 2 videos of the same dance done by different animals to get 3D points.  Modern AI movie generators get 90% of the way to useful output but nothing close to the required fidelity.

Openpose with 2 cameras & a lion could probably go a long way to making all the animation keyframes required.