Reproduce “PE-OPSD: Internalizing Prompt Enhancement into Flow-matching Models via On-Policy Self-Distillation” 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
Mingfeng Lin, Chengfei Cai, Lin Xu, Chengqian Ma, Yuxiang Wei, Liang Han
PE-OPSD transfers prompt enhancement into training for flow-matching text-to-image models. It distills an enhanced-prompt teacher at states reached by a student generating from the original prompt, so inference needs no rewriter. Across three backbones, it leads aggregate prompt-fidelity improvement over Base and improves aggregate visual appeal versus Base, though not every prompt improves or visual appeal always exceeds inference-time enhancement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.