Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation
ATAS uses shared scene and event memories to keep generated shots consistent when revisiting places or replaying earlier events.
The framework turns a screenplay into shot requests with explicit location, camera and event-time references. Before generating each clip, it retrieves stored geometry and time-indexed observations; afterward, it admits new observations to memory only if review and state checks pass. Across four video generators, the authors report improvements on every measured cross-shot consistency metric and the highest aggregate benchmark score for each backbone. ATAS also received the highest human ratings on all six evaluation criteria.
HF Daily Papers' note
The framework turns a screenplay into shot requests with explicit location, camera and event-time references. Before generating each clip, it retrieves stored geometry and time-indexed observations; afterward, it admits new observations to memory only if review and state checks pass. Across four video generators, the authors report improvements on every measured cross-shot consistency metric and the highest aggregate benchmark score for each backbone. ATAS also received the highest human ratings on all six evaluation criteria.
HF Daily Papers' note
score 5