Oneira

From Open-Ended Generation to Open-World Interaction in Video World Models

Oneira: From Open-Ended Generation to Open-World Interaction in Video World Models

Xindi Yang, Baolu Li, Liam Lee, Zhenfei Yin, Songxin Zhang, Zhuoyang Song, Xu Jia, Jianfei Cai, Tien-Tsin Wong, Bingyi Jing, Mengyue Yang† († corresponding author)

Oneira is an interactive video world model built around an explicit world state that a coding agent manages. The agent adds objects that appear during exploration to a world state table and writes the outcome of every interaction into it. An engine renders the updated table into a conditioning video, and the video model generates the next segment from it. As a result, newly generated objects become interactive, and their changed states persist over long horizons.

Code is still under preparation. All code will be available later.

Citation

@misc{yang2026oneira,
  title={Oneira: From Open-Ended Generation to Open-World Interaction in Video World Models},
  author={Xindi Yang and Baolu Li and Liam Lee and Zhenfei Yin and Songxin Zhang and Zhuoyang Song and Xu Jia and Jianfei Cai and Tien-Tsin Wong and Bingyi Jing and Mengyue Yang},
  year={2026},
  eprint={2610.01614},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2610.01614},
}