AI RESEARCH
AdaState: Self-Evolving Anchors for Streaming Video Generation
arXiv CS.CV
•
ArXi:2605.30349v1 Announce Type: new Autoregressive video diffusion models generate streaming video by producing frames sequentially, conditioning each chunk on previously generated content. These models are structurally anchored to the first frame: its key-value representation occupies a privileged position in the attention cache and serves as the primary scene reference throughout generation.