LaWa: Using Latent Space for In-Generation Image Watermarking

2024-08-11

European Conference on Computer Vision

AI Summary

To address the growing risk of misuse of AI-generated images, this paper proposes a robust, imperceptible watermarking method operating directly in the latent space—specifically co-designed with the generation process of Latent Diffusion Models (LDMs). The method leverages a pre-trained autoencoder’s latent space to construct a coarse-to-fine two-stage embedding module and integrates differentiable watermark injection into the LDM’s forward sampling procedure, enabling end-to-end joint optimization. Compared to prior approaches, it achieves superior robustness against common distortions—including cropping, compression, and filtering—while preserving high perceptual quality. It attains a false detection rate below 0.5%, incurs low computational overhead, and generalizes effectively to generic image watermarking tasks. The core contribution lies in a unified generative-watermarking framework grounded in latent-space modeling and co-optimization, marking the first work to tightly couple watermark embedding with LDM-based synthesis.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

With generative models producing high quality images that are indistinguishable from real ones, there is growing concern regarding the malicious usage of AI-generated images. Imperceptible image watermarking is one viable solution towards such concerns. Prior watermarking methods map the image to a latent space for adding the watermark. Moreover, Latent Diffusion Models (LDM) generate the image in the latent space of a pre-trained autoencoder. We argue that this latent space can be used to integrate watermarking into the generation process. To this end, we present LaWa, an in-generation image watermarking method designed for LDMs. By using coarse-to-fine watermark embedding modules, LaWa modifies the latent space of pre-trained autoencoders and achieves high robustness against a wide range of image transformations while preserving perceptual quality of the image. We show that LaWa can also be used as a general image watermarking method. Through extensive experiments, we demonstrate that LaWa outperforms previous works in perceptual quality, robustness against attacks, and computational complexity, while having very low false positive rate. Code is available here.

Problem

• Prevent malicious use of AI-generated images via watermarking
• Integrate watermarking into latent space of diffusion models
• Enhance robustness against image transformations while preserving quality

Innovation

• Latent space watermarking for generative models
• Coarse-to-fine watermark embedding modules
• Robust against image transformations