1 paper
Hongzhou Zhu, Min Zhao, Guande He +3
To achieve real-time interactive video generation, current methods distill pretrained bidirectional video diffusion models into few-step autoregressive (AR) models, facing an archi…