1 paper
Haocheng Xi, Yiming Xie, Hexu Zhao +8
Video diffusion models repeatedly process long spatiotemporal token sequences during denoising, making attention a major computational bottleneck. Linear attention offers an appeal…