Resources / 03
Read it.
Run it. Cite it.
The primary paper, project page and code are the source of truth. This page collects the links and practical requirements in one place.
Primary sources
Start with the work itself.
Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation
Abstract, PDF, HTML, authors, submission history and citation metadata.
ATAS project page
Promo video, pipeline overview, qualitative results, benchmark tables and technical details.
ziplab/ATAS on GitHub
Client, geometry service, examples, design notes, tests and setup instructions.
Run the reference implementation
The public code has two parts.
The client coordinates screenplay compilation and generation. The geometry service reconstructs and registers observations. The repository says the client needs no GPU, while the geometry service needs an NVIDIA GPU with about 20GB free memory.
pip install -r geometry/environment.snapshot.txt
pip install -e geometry
cp geometry/.env.example geometry/.env
cd geometry && ./run_geometry_service.shpip install -e client
cp client/.env.example client/.env
atas run --screenplay examples/bookstore/screenplay.txt --story-id bookstoreIncluded examples
Stories built to test memory.
The repository includes bookstore, clockshop and record examples. The project gallery also follows longer stories such as a laundry day, a winter cabin fire and a bath filling while the character moves between rooms.
Browse the examples ↗Citation
Give the work a stable reference.
Use the paper's BibTeX entry when you build on the method or link to this guide.
@article{wang2026atas,
title = {Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation},
journal = {arXiv preprint arXiv:2610.20571},
year = {2026}
}License and attribution
Reuse with context.
The ATAS repository is released under the MIT License, while third-party code in its geometry backbones keeps its own license files. Check the repository before redistributing code or model weights.
Read the license ↗