I'm not sure exactly what you're asking, but there are controlnet models that can be made from videos or images and used to guide how a compatible generative AI model poses figures, and there may be something that does functionally the same thing for trying to animate bones for a 3d model although I have to stress that a lot of this tech is at once surprisingly capable but still complete dogshit in practice. Generative AI research has been hard focused on making the shitty little black boxes less bad at churning out slop from simple prompts in a way that's meant a whole bunch of attendant tech that might make it less bad via human curation and guidance has just not been made.

The hobbyist sector wants its slop gacha, the management sector wants a fully autonomous worker replacer, and the whole thing's such a grift no thought's been given to how to actually make genuinely useful toolkits involving the tech.

[–] ZWQbpkzl@hexbear.net 2 points 2 weeks ago

I meant gen-ai that made an existing 3D move naturally, which is what you described.

Absolutely agreed that gen-ai is best suited as a smaller tool in a makers tool set. But you can't make it big selling tools like that. Basically anyone can compete with you and a free open-source model will probably win. You can only justify massive investments if the product will be scarce and that's done by requiring these massive data centers to rub huge models.