Zhejiang et al. open-source ATAS: screenplay-to-film with persistent 3D world state
lmoroney · x · 2026-10-16
A team from Zhejiang University, Monash, UC Berkeley and Adelaide open-sourced ATAS (Any Time, Any Space), a world-state-centric framework for agentic video generation.
- Problem: AI-generated shorts break spatial consistency between shots (e.g., furniture silently rearranging).
- Approach: Takes a plain-text screenplay and maintains a running world state behind the film. A scene memory stores each location's geometry in one shared 3D frame; every shot is generated by projecting the world state into the requested view, then reconstructed, registered and validated before updating the world.
- Core piece: a World-State Compiler that turns a screenplay into a bounded world description.
MIT-licensed on GitHub, already at 203 stars.
More from Multimodal
- Shanghai AI Lab unveils Intern Lumina U2, a diffusion LLM unifying image, video and 3D tasks — Shanghai-AI-Laboratory · 2026-10-16
- NAVER AI Lab proposes Speculative Forcing to speed up autoregressive video generation — naver-ai · 2026-10-16
- MFVSR: A Multi-Style Benchmark with 3,000 Videos and 16 Languages for Video Subtitle Removal — Xiyan Huang · 2026-10-16
- OmniCompass: PolyU launches a holistic benchmark for omni-modal video grounding and QA — VCLab-HKPU · 2026-10-16
- Researcher admits his prediction that text-to-image would incubate AI control research was falsified — jd_pressman · 2026-10-16
- AI turns Kuroko's Basket rivals into muscle-bound shoppers in viral clip — EternalSnow05 · 2026-10-16