Few-Step Diffusion Sampling Through Instance-Aware Discretizations
Liangyu Yuan 1,2 Ruoyu Wang 1 Tong Zhao 1,3 Dingwen Fu 1 Mingkun Lei 1 Beier Zhu 4 Chi Zhang 1† 1 Westlake University 2 Tongji University 3 Zhejiang University 4 Nanyang Technological University · 2026
Read Few-Step Diffusion Sampling Through Instance-Aware Discretizations by Liangyu Yuan 1,2 Ruoyu Wang 1 Tong Zhao 1,3 Dingwen Fu 1 Mingkun Lei 1 Beier Zhu 4 Chi Zhang 1† 1 Westlake University 2 Tongji University 3 Zhejiang University 4 Nanyang Technological University (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-2603-17671.txt.
Description
Yuan and co-authors speed up diffusion sampling by tailoring step schedules to each input. Most solvers use one global timetable for all samples. Controlled tests show that choice wastes steps on easy cases and starves hard ones. Their framework learns timestep splits from input-dependent priors. It extends gradient-based schedule search to conditional generation. Gains held across images, latent models, and video flow matching, they report, at near-zero run cost.
Catalog record
| CREATOR | Liangyu Yuan 1,2 Ruoyu Wang 1 Tong Zhao 1,3 Dingwen Fu 1 Mingkun Lei 1 Beier Zhu 4 Chi Zhang 1† 1 Westlake University 2 Tongji University 3 Zhejiang University 4 Nanyang Technological University |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61138 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.17671