The core move

Same diffusion, one more dimension.

Video models denoise a block of spacetime — width × height × time — so every frame is generated aware of its neighbours, not stitched together afterwards.