Background
Generative AI has rapidly transformed how text, images and multimedia are created, enabling anyone to produce high-quality content with minimal effort. However, this democratization also makes it easy to misuse or misappropriate copyrighted material. Embedding robust, invisible watermarks directly into model outputs offers a practical way for rights holders to assert ownership, even when they have no access to the model’s internals.
Challenges
Demo
![[Picture]](https://www.owlstown.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsiZGF0YSI6NDQ0NjE0LCJwdXIiOiJibG9iX2lkIn19--41abff9c9d90f094ee0108e7ea333854d60774ed/fig-deemo.gif){kind=link}
Related Publications
The Latent Seal: Robust Model Watermarking for Latent Diffusion Model
Qin Zhao, Tong Liu, Xiaochen Yuan, Guoheng Huang, Xueyuan Gong, Wei Wang
Machine Intelligence Research
The Latens Patronus: Seamless Model Watermarking for Latent Diffusion Model in IoT Environments
Qin Zhao, Tong Liu, Wei Ke, Guoheng Huang, Xueyuan Gong, Xiaochen Yuan
IEEE Internet of Things Journal, 2026 May 25, pp. 1-1