This paper shows how to make long, camera-controlled videos much faster by generating only a few smart keyframes with diffusion, then filling in the rest using a 3D scene and rendering.
PlenopticDreamer is a new way to remake a video from different camera paths while keeping everything consistent across views and over time.