What / who is the leading edge of 3D AI? I took a look into it last year and the best tools were still a very long away from producing usable results (couldn't keep tri count under control, spiderweb rigging, shape jank, etc).
Still super early days though, I did have some of the same issues you had.
I'd say it'll take them some years to finish figuring out 3D, it's a good stretch beyond 2D image generation.