Codec guided watermarking improves traceability of diffusion images
CLaW: Codec-Guided Adaptive Latent Watermarking for Traceable Diffusion Image Generation
Computer Vision and Pattern RecognitionCryptography and Security
Summary
AI can now create very realistic images from text, but this makes it hard to tell which images were made by which user, causing issues like misuse and copyright problems. The authors present CLaW, a method that hides special codes inside generated images without changing how they look much. These codes can be detected later without needing to change the AI model, even after the image is edited. This helps track who created the image while keeping pictures clear and resistant to distortions.
What this means in practice
- •For digital content platforms: Enable tracking of AI-generated images per user to help prevent misuse and support copyright enforcement.
- •For image editing software developers: Integrate watermark detection that remains effective after common image transformations to authenticate images reliably.
Authors
Huangchao Song, Lingyun Yu, Cong Li, Peiqi Jiang, Mingqi Fang, Hongtao Xie
Abstract
With the rapid advancement of text-to-image diffusion models, increasingly realistic AI-generated content has raised serious concerns about misuse and copyright infringement. Digital watermarking offers a promising solution by enabling user-level traceability, yet existing methods remain difficult to deploy at scale, hindered by costly model retraining or detection, poor effectiveness-fidelity trade-offs, and limited robustness to image transformations. To address these challenges, we propose CLaW (Codec-guided Latent Watermarking), a robust and efficient watermarking framework for traceable diffusion image generation with frozen diffusion backbones and inversion-free detection. Specifically, CLaW first uses a pretrained watermark codec to map each watermark message to a latent residual, enabling low-cost sampling-time injection. Then, the residual is injected within a late denoising window, where image semantics are better preserved and the watermark signal is less perturbed by subsequent denoising updates, yielding a better balance between watermark effectiveness and image fidelity. Furthermore, we introduce a decoder-guided adaptive injection mechanism that uses decoding confidence as feedback to dynamically calibrate watermark strength, reinforcing weak watermark signals and improving robustness under image transformations. Extensive experiments show that CLaW preserves high visual fidelity and achieves state-of-the-art robustness under image transformations, enabling reliable user-level traceability.