English

Robustness of Watermarking on Text-to-Image Diffusion Models

Cryptography and Security 2024-11-05 v2

Abstract

Watermarking has become one of promising techniques to not only aid in identifying AI-generated images but also serve as a deterrent against the unethical use of these models. However, the robustness of watermarking techniques has not been extensively studied recently. In this paper, we investigate the robustness of generative watermarking, which is created from the integration of watermarking embedding and text-to-image generation processing in generative models, e.g., latent diffusion models. Specifically, we propose three attacking methods, i.e., discriminator-based attacks, edge prediction-based attacks, and fine-tune-based attacks, under the scenario where the watermark decoder is not accessible. The model is allowed to be fine-tuned to created AI agents with specific generative tasks for personalizing or specializing. We found that generative watermarking methods are robust to direct evasion attacks, like discriminator-based attacks, or manipulation based on the edge information in edge prediction-based attacks but vulnerable to malicious fine-tuning. Experimental results show that our fine-tune-based attacks can decrease the accuracy of the watermark detection to nearly 67.92%67.92\%. In addition, We conduct an ablation study on the length of fine-tuned messages, encoder/decoder's depth and structure to identify key factors that impact the performance of fine-tune-based attacks.

Keywords

Cite

@article{arxiv.2408.02035,
  title  = {Robustness of Watermarking on Text-to-Image Diffusion Models},
  author = {Xiaodong Wu and Xiangman Li and Jianbing Ni},
  journal= {arXiv preprint arXiv:2408.02035},
  year   = {2024}
}

Comments

We find an error in one of the proposed attack methods, which significantly impact the correctness. In addition, the experiment is not solid enough to support the results

R2 v1 2026-06-28T18:03:30.756Z