>>720615969
I really really hope that a parallel approach will start getting worked on - a model that is actually sensitive to 3d space instead of just being trained on 2d images. This concept of "give a prompt in a game and it just happens," with the current form of AI, can only generate video originating from an initial image and unavoidably deviates more and more as time passes. Instead train the AI on 3d modeling, texturing and animations and it can generate in-engine scenes for you.