Gaussian Shannon: High-Precision Diffusion Model Watermarking Based on Communication

2026-03-27

AI Summary

This work addresses the limitations of existing diffusion model watermarking methods, which rely on threshold-based detection and support only fuzzy matching, thereby hindering bit-exact recovery of structured watermarks and restricting offline verification and lossless metadata applications. The authors model the diffusion process as a noisy communication channel and embed watermarks directly into the initial Gaussian noise without requiring model fine-tuning or compromising image quality. By integrating error-correcting codes with a majority voting mechanism, they propose the first framework enabling lossless embedding and precise bit-level recovery of structured watermarks. A cascaded defense strategy is further designed to counter both localized bit flips and global distortions. Evaluated across three Stable Diffusion variants and seven perturbation types, the method achieves state-of-the-art bit accuracy and true positive rates, significantly enhancing the reliability of copyright attribution.

Abstract

Diffusion models generate high-quality images but pose serious risks like copyright violation and disinformation. Watermarking is a key defense for tracing and authenticating AI-generated content. However, existing methods rely on threshold-based detection, which only supports fuzzy matching and cannot recover structured watermark data bit-exactly, making them unsuitable for offline verification or applications requiring lossless metadata (e.g., licensing instructions). To address this problem, in this paper, we propose Gaussian Shannon, a watermarking framework that treats the diffusion process as a noisy communication channel and enables both robust tracing and exact bit recovery. Our method embeds watermarks in the initial Gaussian noise without fine-tuning or quality loss. We identify two types of channel interference, namely local bit flips and global stochastic distortions, and design a cascaded defense combining error-correcting codes and majority voting. This ensures reliable end-to-end transmission of semantic payloads. Experiments across three Stable Diffusion variants and seven perturbation types show that Gaussian Shannon achieves state-of-the-art bit-level accuracy while maintaining a high true positive rate, enabling trustworthy rights attribution in real-world deployment. The source code have been made available at: https://github.com/Rambo-Yi/Gaussian-Shannon

Problem

• diffusion model
• watermarking
• bit-exact recovery
• structured watermark
• offline verification

Innovation

• diffusion watermarking
• bit-exact recovery
• communication channel modeling
• error-correcting codes
• Gaussian noise embedding