I suspect you are overthinking it, or at least should acknowledge that one needs to take the idea and walk a long, long way!

But, with that said, this work is dependent on having a "source" that the user is using as an input for pose detection. The actual accomplishment must still be performed and recorded, though I suppose this opens the door to the dance equivalent of "lip syncing" even beyond what might be done today with a body double.