Reproduce “ZEUS: Accelerating Diffusion Models with Only Second-Order Predictor” 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
Yixiao Wang, Ting Jiang, Zishan Shao, Hancheng Ye, Jingwei Sun, Mingyuan Ma, Jianyi Zhang, Yiran Chen and 1 more
ZEUS reduces denoiser evaluations during diffusion and flow sampling by estimating skipped outputs from the latest computed output and its change, then alternating estimates with reuse of that output. On FLUX.1-dev text-to-image, ZEUS-Medium reaches 2.09× speedup, LPIPS 0.047, and FID 1.53, versus SADA’s 2.02×, 0.060, and 1.95; quality is not unchanged at every speed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.