#spatiotemporal modeling

Alibaba introduces Lumos-1, a unified autoregressive video generation model leveraging MM-RoPE and AR-DF to efficiently model spatiotemporal dependencies, achieving competitive results with efficient training.