Reproduce “ACEdit: Stable Multi-Turn Image Editing via Asymmetric Conditioning” 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
Sara Dorfman, Nir Goren, Gal Barak, Daniel Cohen-Or, Or Patashnik
ACEdit trains image editors on self-generated multi-turn histories. The student edits its previous output, while a frozen teacher evaluates results against the original image and accumulated instructions. On EdiVal-10, ACEdit scores 3.85 for consistency versus 3.03 for the base model, with instruction-following scores of 3.38 and 3.31, respectively. This supports better preservation in that evaluation, not guaranteed success on difficult edits.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.