1 paper
Taewoo Ha, Shafayat Mowla Anik, Dae Yeol Lee +2
Video diffusion transformers depend on temporal attention to coordinate information across frames, yet nearly everything known about this mechanism comes from analyzing trained mod…