Reproduce “AnimWeb-Bench: Benchmarking Animated Webpage Code Generation from Videos” 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

Zhongyi Zhang, Jingyu Xiao, Shuqing Li, Michael R. Lyu

AnimWeb-Bench tests page-load animation reconstruction by asking models to generate webpage code from reference videos and original assets. It measures settled appearance separately from motion between states. Across evaluated models, every animation metric is lower than every static metric, though the dimensions are not directly comparable. Codex inspection and render feedback improve fidelity, especially motion, in this benchmark, but remain diagnostic interventions, not tuned methods.

Summary by alphaXiv.

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