ZEUS: Accelerating Diffusion Models with Only Second-Order Predictor
Yixiao Wang, Ting Jiang, Zishan Shao, Hancheng Ye, Jingwei Sun, Mingyuan Ma, Jianyi Zhang, Yiran Chen, Hai Li · 2026
Read ZEUS: Accelerating Diffusion Models with Only Second-Order Predictor by Yixiao Wang, Ting Jiang, Zishan Shao, Hancheng Ye, Jingwei Sun, Mingyuan Ma, Jianyi Zhang, Yiran Chen, Hai Li (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-01552.txt.
Description
Wang and colleagues propose ZEUS, a method predicting fewer denoiser runs with a second-order model, stabilizing aggressive skipping with interleaved steps avoiding back-to-back extrapolations. Beyond twofold acceleration, step skipping leaves at most one fresh run per window, so the computed output and its backward difference are the only grounded facts. ZEUS adds zero overhead, no feature caches, and no design changes. Across image and video generation, ZEUS improves speed-fidelity results over recent training-free baselines, with up to 3.2 times end-to-end speedup while keeping quality.
Catalog record
| CREATOR | Yixiao Wang, Ting Jiang, Zishan Shao, Hancheng Ye, Jingwei Sun, Mingyuan Ma, Jianyi Zhang, Yiran Chen, Hai Li |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 94443 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.01552