A Somewhat Robust Image Watermark against Diffusion-based Editing Models

TL;DR

This paper introduces RIW, a novel adversarial example-based invisible watermarking method that remains robust against diffusion model-based image editing, achieving 96% watermark extraction accuracy.

Contribution

The paper presents the first formalization of watermarking challenges against diffusion-based editing and proposes RIW, a new technique that significantly improves watermark robustness in this context.

Findings

RIW achieves 96% watermark extraction accuracy after editing.

Traditional watermarking methods fail to extract watermarks post-editing.

The proposed method outperforms conventional techniques in robustness against diffusion-based edits.

Abstract

Recently, diffusion models (DMs) have become the state-of-the-art method for image synthesis. Editing models based on DMs, known for their high fidelity and precision, have inadvertently introduced new challenges related to image copyright infringement and malicious editing. Our work is the first to formalize and address this issue. After assessing and attempting to enhance traditional image watermarking techniques, we recognize their limitations in this emerging context. In response, we develop a novel technique, RIW (Robust Invisible Watermarking), to embed invisible watermarks leveraging adversarial example techniques. Our technique ensures a high extraction accuracy of $96\%$ for the invisible watermark after editing, compared to the $0\%$ offered by conventional methods. We provide access to our code at https://github.com/BennyTMT/RIW.