Conclusion seems to be that the models are currently very good at doing static environments and machines, including machine animation, bad at doing organic beings and animations related to them. When asked, they say as much, and it's because "correctness is externally verifiable," i.e. it's easy to mechanically verify that a robot's gait is correct, not so much for a human's animation which requires perceptual judgement. Lots of game possibilities unlocked for me personally, since I never spent time focusing on learning 3D pipelines before, a good decision in retrospect, since it's all so easy now, but truly, it's now primarily a matter of having good ideas and being able to focus on seeing them through completion, which, in a way, has always been the case.