Reproduce “MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion” 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

Houyuan Chen, Fanzhou Wang, Xiangyu Fan, Zesong Qiu, Yukun Wei, Songlin Yang, Zhitao Yang, Lvmin Zhang and 5 more

MT Forcing trains multi-turn image editors to use their own outputs, matching them to a single-turn teacher while adding supervision on reference editing histories. On the eight-turn MTEdit-Bench, SenseNova-U1.5 with MT Forcing scores 9.17 versus 4.52 on instruction following and 7.68 versus 3.22 on content consistency; Pixel displacement is 0.71 versus 6.89. Results are limited to this benchmark.

Summary by alphaXiv.

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