Reproduce “Enabling Creative Exploration for Vibe Design Agents” 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
Yifan Zhang, Nghi D. Q. Bui, Georgios Evangelopoulos, Arnaud Benard
The system adds a proposal-and-selection stage for choosing a theme or image prompt before implementation, while keeping token-decoding settings fixed. On the 83-prompt benchmark, theme sampling at τ=2.0 raised option-label coverage from 1.00 to 2.87 and lowered screenshot similarity from 0.6765 to 0.5438. Asset sampling had a 1.32 LLM-judge win/loss ratio; these results do not establish human preference or functionality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.