ziplab/ATAS

Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation Weijie Wang1,   Haomin Zhang1,   Ye Li1,   Akide Liu2   Zeyu Zhang3 Haoyu Zhao1   Yefei He1   Donny Y. Chen2   Feng Chen4   Bohan Zhuang1,† 1Zhejiang University     2Monash University     3BAIR, UC Berkeley     4Adelaide University Equal contribution.   †Corresponding author. ATAS keeps one persistent world behind a whole film. Every shot is generated by projecting the world state into the requested view, and every generated shot is reconstructed, registered and validated before it updates that world. A world-state compiler turns the screenplay into shot requests; each request queries the scene and event memory, the rendered context conditions the video generator, and the new clip is reconstructed and registered before it can enter the world. Human evaluation against multi-shot baselines (1–5). Spatial revisit and event recapture. Runtime, candidate lifecycle and delivery…