Scalable Adaptive Computation for Iterative Generation

Presents RINs, which yields SotA image and video generation without cascades or guidance, while being domain-agnostic and up to 10x more efficient compared to specialized 2D and 3D U-Nets.

Implement RINs to achieve state-of-the-art image and video generation without the need for cascades or guidance, and with improved efficiency compared to traditional U-Nets.