Luminark: Training-free, Reliable Watermarking for General Vision Generative Models

Jiayi Xu, Zhang Zhang, Yuanrui Zhang, Ruitao Chen, Yixian Xu, Tianyu He, Di He

Keywords: security, generative model, watermark

Abstract: Watermarking is a fundamental technique for protecting digital visual content. However, developing a general and reliable watermarking method for vision generative models remains an open challenge due to the diversity of generative paradigms and design choices. In this paper, we introduce \emph{Luminark}, a training-free, robust and general watermarking method for vision generative models. Our approach is built upon a novel watermark definition that leverages patch-level luminance statistics. Specifically, the service provider predefines a binary pattern together with corresponding patch-level thresholds. To detect a watermark in a given image, we evaluate whether the luminance of each patch surpasses its threshold and then verify whether the resulting binary pattern aligns with the target one. A simple statistical analysis demonstrates that the false positive rate of the proposed method can be effectively controlled, thereby ensuring reliable detection. To enable seamless watermark injection across different paradigms, we leverage the widely adopted guidance technique as a plug-and-play mechanism and develop the \emph{watermark guidance}. This design enables Luminark to achieve generality across state-of-the-art generative models without compromising image quality. Empirically, we evaluate our approach on nine models spanning diffusion (EDM2 family), autoregressive (VAR family), and hybrid (MAR family) frameworks. Across all evaluations, Luminark consistently demonstrates high detection accuracy, strong robustness against common image transformations, and good performance on visual quality.

Primary Area: generative models

Submission Number: 2300