I could see a system where you very loosely define a handful of keyframes, give it some kind of description for what your looking for, and then the model fills in the rest for you. It almost certainly would still require some tweaking on the artists part, but I think a huge benefit is that AI can be non-deterministic. If you don’t like what you got you can just run it again without changing inputs. Maybe save multiple runs and manually combine all the best elements. Lots of opportunity.