T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models
Jindong Yang, Han Fang, Weiming Zhang, Nenghai Yu, Kejiang Chen · 2025
Read T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models by Jindong Yang, Han Fang, Weiming Zhang, Nenghai Yu, Kejiang Chen (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-2510-22366.txt.
Description
Jindong Yang and four co-authors tackle a trade off in watermarking images made by diffusion models. Noise based methods hide a mark in the starting noise, but strong marks can limit image variety, while looser marks can be too weak to survive. Their answer is a two stage scheme called T2SMark.
T2SMark places bits only in the reliable tail parts of the noise and samples the middle part at random. A random session key is mixed into both stages to keep outputs varied. Tests on models with two kinds of backbones show a better balance of strength and variety.
Catalog record
| CREATOR | Jindong Yang, Han Fang, Weiming Zhang, Nenghai Yu, Kejiang Chen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-22366