PhaseMark: A Post-hoc, Optimization-Free Watermarking of AI-generated Images in the Latent Frequency Domain
This addresses the need for efficient and robust watermarking to combat the proliferation of hyper-realistic images from Latent Diffusion Models, with potential applications in content verification and copyright protection.
The paper tackled the problem of slow post-hoc watermarking for AI-generated images by introducing PhaseMark, a single-shot, optimization-free method that modulates phase in the VAE latent frequency domain, resulting in thousands of times faster speed and state-of-the-art resilience against severe attacks without quality degradation.
The proliferation of hyper-realistic images from Latent Diffusion Models (LDMs) demands robust watermarking, yet existing post-hoc methods are prohibitively slow due to iterative optimization or inversion processes. We introduce PhaseMark, a single-shot, optimization-free framework that directly modulates the phase in the VAE latent frequency domain. This approach makes PhaseMark thousands of times faster than optimization-based techniques while achieving state-of-the-art resilience against severe attacks, including regeneration, without degrading image quality. We analyze four modulation variants, revealing a clear performance-quality trade-off. PhaseMark demonstrates a new paradigm where efficient, resilient watermarking is achieved by exploiting intrinsic latent properties.