Reproduce “Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network” 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
Yulong Chen, Ziqian Zhang, Haoyu Zhang, Ao He, Yaxing Wang, Senmao Li, Kai Wang
RefineEdit edits prompt-generated images during refinement, using probability differences between source and edit branches to select bits for change while preserving source bits. Adaptive spatial freezing and finite bit locking stabilize this routing. On PIE-Bench, it leads evaluated methods on unedited-content metrics and prompt-alignment scores, though FlowEdit leads Structure Distance. These results do not establish performance on arbitrary real photographs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.