Reproduce “WorldCast: Exploring Intelligent Visual Generation in World State Prediction” 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
Wenhao Yan, Haodong Li, Xiaotong Li, Yaxin Luo, Zhengrong Yue, Jialv Zou, Xu Huang, Haojin Yang and 4 more
WorldCast evaluates whether systems can generate ordered visual states from instructions and references. It covers 320 cases across eight categories, requesting 1,352 states. Systems are scored on task completion, causal plausibility, and visual consistency. GPT-Image-2.5 Sunburst paired with GPT-6 Astra scores 72.8, versus 47.2 for Seedance 2.0. These scores measure benchmark sequences, not internal world models or video quality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.