REG: Rectified Gradient Guidance for Conditional Diffusion Models
Zhengqi Gao, Kaiwen Zha, Tianyuan Zhang, Zihui Xue, Duane S. Boning · 2025
Read REG: Rectified Gradient Guidance for Conditional Diffusion Models by Zhengqi Gao, Kaiwen Zha, Tianyuan Zhang, Zihui Xue, Duane S. Boning (2025) 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-2501-18865.txt.
Description
Zhengqi Gao and co-authors fixed the math beneath diffusion guidance in two thousand twenty-five. Guidance steers image generators toward a prompt. Its practice works, yet its theory was shaky. The team proves the usual target distribution is invalid. A scaled joint distribution is the right target instead. Existing methods approximate an intractable ideal blindly.
Rectified gradient guidance follows from the cleanup. It upgrades current guidance techniques as a plug-in. Toy one- and two-dimensional tests came first. Rectified guidance tracked the true optimum closer. Better theory promises better-behaved generators. Conditional sampling is everywhere in creative tools. The paper gives practitioners a firmer footing.
Catalog record
| CREATOR | Zhengqi Gao, Kaiwen Zha, Tianyuan Zhang, Zihui Xue, Duane S. Boning |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 82 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-18865