Reproduce “AgiWorld: An Agentic Video World Model for Building Persistent Worlds” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shaoheng Fang, Yiqun Mei, Hanwen Jiang, Xin Sun, Haitian Zheng, Hao Tan, Kalyan Sunkavalli, Georgios Pavlakos and 2 more
AgiWorld takes one image and a free-form instruction to plan a journey through an imagined world while generating video in chunks. An agent revises its plan after each chunk, using spatial memory that selects surfaces visible from the new camera and maintains reliable geometry. On World Explore, AgiWorld-WM reports FID 35.37 versus Lyra 2.0’s 42.73, for tested 60-second camera paths.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.