Reproduce “WM^MLLM: Turning MLLMs into World Models” 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
Baiang Li, Enze Xie, Jincheng Yu, Jin Wang, Haozhe Liu, Yitong Li, Juanxi Tian, Felix Heide
WM^MLLM stores a panorama and revision history; an MLLM plans edits, image generation proposes views, and video presents transitions. It supports saved-state revisits and local updates. On MBench-A, it completes 85.3% of camera round trips, versus 50.9% for HY-WorldPlay. Post-generation region selection raises judged completion from 34.8% to 83.0% across 112 edits. Results concern a one-camera-center panorama and imperfect verification.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.