✦
Summary
ATAS, short for Any Time, Any Space, is an agentic framework for long-form video with a shared spatial and temporal world state. It converts a screenplay into shot requests with location, camera, and event-time references, retrieving relevant evidence before generating a shot.
Key Features
Screenplay-to-shot compilation.
Explicit location references.
Camera-aware shot requests.
Event-time indexing.
Shared world-state memory.
Evidence retrieval.
Nonconsecutive shot generation.
Long-form continuity.
About ATAS
Creators and researchers can revisit locations or earlier events without relying only on recent video frames to preserve continuity. Core capabilities include Screenplay-to-shot compilation, Explicit location references, Camera-aware shot requests, and Event-time indexing.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Shared world-state memory, Evidence retrieval, Nonconsecutive shot generation, and Long-form continuity.