Reproduce “ScaleWorld: High-Resolution World Generation through Shared Prediction and Refinement” 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

Yansong Qu, Wesker Zheng, Quanjian Song, Zhiyuan Zhang, Xiao Cao, Shengchuan Zhang, Liujuan Cao

ScaleWorld generates camera-controlled visual worlds chunk by chunk, using one causal backbone to predict a low-resolution chunk and refine it into high resolution. It keeps the prediction usable and maintains separate histories. For complete generation, it scores 81.60 on WBench Average, compared with 81.58 for a separate specialist cascade; these comparable scores do not establish physical-state simulation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.