Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PhaseMark: A Post-hoc, Optimization-Free Watermarking of AI-generated Images in the Latent Frequency Domain

Sung Ju Lee, Nam Ik Cho

TL;DR

The paper tackles robust watermarking for AI-generated images from Latent Diffusion Models, where existing post-hoc methods are slow due to optimization or inversion. PhaseMark offers a single-shot, optimization-free solution that modulates phase in the mid-band of the VAE latent frequency domain to embed an $L$-bit payload. It presents four variants—APM, PCQ, IPS, and SPS—exhibiting a clear trade-off between detection performance and image quality, with axis-offsets further boosting fidelity. Compared with baselines, PhaseMark achieves state-of-the-art resilience to regeneration and cropping while maintaining real-time embedding/detection and minimal $\\Delta$FID-1k and thus enables scalable, post-hoc watermarking for generative AI; cryptographic payload protection (e.g., ChaCha20) is discussed as a security extension.

Abstract

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.

PhaseMark: A Post-hoc, Optimization-Free Watermarking of AI-generated Images in the Latent Frequency Domain

TL;DR

The paper tackles robust watermarking for AI-generated images from Latent Diffusion Models, where existing post-hoc methods are slow due to optimization or inversion. PhaseMark offers a single-shot, optimization-free solution that modulates phase in the mid-band of the VAE latent frequency domain to embed an -bit payload. It presents four variants—APM, PCQ, IPS, and SPS—exhibiting a clear trade-off between detection performance and image quality, with axis-offsets further boosting fidelity. Compared with baselines, PhaseMark achieves state-of-the-art resilience to regeneration and cropping while maintaining real-time embedding/detection and minimal FID-1k and thus enables scalable, post-hoc watermarking for generative AI; cryptographic payload protection (e.g., ChaCha20) is discussed as a security extension.

Abstract

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.
Paper Structure (13 sections, 4 equations, 2 figures, 3 tables)

This paper contains 13 sections, 4 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Latent Phase Watermarking
  3. Overall Framework
  4. Phase Modulation for Embedding
  5. Absolute Phase Modulation
  6. Relative Phase Modulation
  7. Watermark Detection
  8. Experiments
  9. Experimental Setup and Implementation Details
  10. Comparison with Baselines
  11. Analysis of Proposed Variants
  12. Ablation Study
  13. Conclusion

Figures (2)

  • Figure 1: Performance vs. image quality trade-off for the four proposed variants (APM, PCQ, IPS, SPS). The comparison includes ZoDiac and illustrates the positive effect of applying axis offsets.
  • Figure 2: Generalization performance. PhaseMark maintains consistent detection rates and image quality across three datasets, demonstrating high stability.