Researchers publishing on Hugging Face have introduced Any Time, Any Space (ATAS), a framework aimed at making long-form agentic video generation spatially and temporally consistent across nonconsecutive requests. The paper argues that passing prompts or recent frames between shots is insufficient, since it provides neither a shared spatial frame for revisiting a scene nor a time-indexed record for replaying an earlier event.
ATAS compiles a screenplay into shot requests that carry explicit references to location, camera, and event time. Before generation, the system retrieves relevant evidence from two complementary memories: a scene memory of registered static geometry and an event memory of dynamic observations indexed by scene, event, and time. After generation, ATAS reconstructs and registers the resulting clip.
The framework updates memory only when review and state checks accept an observation, while film delivery remains independent of memory admission. According to the authors, this lets agents revisit earlier locations or replay past events from new viewpoints by querying the same evolving world state.
Across four video generators, ATAS is reported to improve all cross-shot consistency measures and to achieve the highest aggregate benchmark score for each backbone. It also outperforms the compared agents and 3D-guided methods on target-image spatial revisit, leads three event-recapture criteria, and receives the highest human ratings on all six criteria.
Video results and code are available on the project page at lhmd.top/atas/, which includes an interactive demo for exploring point clouds, camera poses, and linked videos.