1 paper · 1 filter
Chenxi Xia, Yixuan Xu, Shan Liu +7
Autoregressive video generation models generate videos sequentially, with each new block conditioned on previously generated blocks. By training on the model's own rollouts, Self F…