Reproduce “BoxWorld: Let World Engines Control and Video Generators Render” 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
Zike Wu, Ahmed Hatem, Chintan Trivedi, Wallace Michel Pinto Lira, Xuebin Qin, Lele Wang, Renjie Liao
BoxWorld separates rule handling from video rendering: an editable engine updates object states, and a generator renders the changes. Its interface gives each object a layout box, persistent identity and appearance, and changing status. On PairedStateBench, it achieves 97.3% Paired-CF, versus at most 16.2% for baselines; both outcomes must succeed in each of 37 paired trials, not all 87 cases.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.