It really feels like for 3D the better quality/compute trade-off would be to have the ML model generate 3D models and animations, and then use a more traditiona

It really feels like for 3D the better quality/compute trade-off would be to have the ML model generate 3D models and animations, and then use a more traditional 3D rendering pipeline (by then with ray tracing and denoising).