DiffMark: Diffusion-based Robust Watermark Against Deepfakes

doi:10.1016/j.inffus.2025.103801

DiffMark: Diffusion-based Robust Watermark Against Deepfakes

2025-10-04

0

PRE

AI

DOI：10.1016/j.inffus.2025.103801

Original

Original request for help

Save

Abstract

En 中文

• Diffusion model is exploited for generative image watermarking. • Cross information fusion module is designed for fusing image and watermark features. • A pre-trained frozen autoencoder is adopted to enhance watermark robustness. • Deepfake-resistant guided diffusion sampling for robust watermarked image generation. • The performance of our model outperforms existing methods.

DiffMark: Diffusion-based Robust Watermark Against Deepfakes

Abstract

Journal

Organization