The problems with full motion video are mostly not light transport problems, they're animation problems. Capturing realistic human movement is tricky enough, l

The problems with full motion video are mostly not light transport problems, they're animation problems. Capturing realistic human movement is tricky enough, let alone generating it, and people are exquisitely sensitive to the creepy effect of unnatural movement. Also, even James Cameron's budget might have been insufficient to model out the material properties of physically credible alternate world, making everything weirdly glossy and fluorescent as an artistic oversight rather than a limitation of the rendering technology. Creating realistic video panning over still-life scenes, like the view out of a helicopter, is a solved problem at sufficient modelling budgets, you don't need AI just fully described scene and lots of flops to run your light transport algorithm. Colorizing NNs are cute but not relevant to this problem. There's a good analogue to sound: a long time ago we mastered algorithms to precisely simulate any musical instrument, acoustic environment, pre-digital audio tranformation, etc. but we still use voice actors because modelling human speech behavior is a much harder problem that modelling audio physics.