AdaState: Self-Evolving Anchors for Streaming Video Generation 事件

AdaState: Self-Evolving Anchors for Streaming Video Generation arXiv:2605.30349v1 Announce Type: new Abstract: 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. As the cleanest and most er