A Watermark for Auto-Regressive Image Generation Models

View on arXiv ← Back to list

Authors: Yihan Wu, Xuehao Cui, Ruibo Chen, Georgios Milis, Heng Huang

Published: 2025-06-13 00:15:54+00:00

AI Summary

This paper proposes C-reweight, a novel distortion-free watermarking method for autoregressive image generation models. C-reweight addresses the "retokenization mismatch" problem by using a clustering-based strategy, improving watermark detectability while preserving image quality.

Abstract

The rapid evolution of image generation models has revolutionized visual content creation, enabling the synthesis of highly realistic and contextually accurate images for diverse applications. However, the potential for misuse, such as deepfake generation, image based phishing attacks, and fabrication of misleading visual evidence, underscores the need for robust authenticity verification mechanisms. While traditional statistical watermarking techniques have proven effective for autoregressive language models, their direct adaptation to image generation models encounters significant challenges due to a phenomenon we term retokenization mismatch, a disparity between original and retokenized sequences during the image generation process. To overcome this limitation, we propose C-reweight, a novel, distortion-free watermarking method explicitly designed for image generation models. By leveraging a clustering-based strategy that treats tokens within the same cluster equivalently, C-reweight mitigates retokenization mismatch while preserving image fidelity. Extensive evaluations on leading image generation platforms reveal that C-reweight not only maintains the visual quality of generated images but also improves detectability over existing distortion-free watermarking techniques, setting a new standard for secure and trustworthy image synthesis.

Key findings

C-reweight significantly outperforms existing distortion-free watermarking techniques in detectability across multiple datasets. It maintains high detection accuracy even under l2 and l∞ noise attacks, demonstrating robustness. The method also preserves image quality as measured by FID and CLIP scores.

Approach

C-reweight uses a clustering-based approach to mitigate retokenization mismatch in watermarking. It groups similar tokens into clusters, treating tokens within the same cluster equivalently during watermark embedding and detection. This allows for reliable watermark detection even after image encoding and decoding.

Datasets

Pickapic_v2, MSCOCO, Flickr8k, LAION

Model(s)

Emu3

Author countries

USA

← Previous