Embedding Watermarks in Stable Diffusion Models

Paper Title:

A Training-Free Plug-and-Play Watermark Framework for Stable Diffusion

Published on:

8 April 2024

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Guokai Zhang,

Lanjun Wang,

Yuting Su,

An-An Liu

Bullets

Key Details

•

Proposes training-free, plug-and-play watermarking framework for Stable Diffusion

•

Embeds watermarks in latent space, adapting to denoising process

•

Achieves balance between image quality and watermark invisibility

•

Shows robustness against various attacks

•

Demonstrates generalization across multiple SD versions

Explore the topics in this paper

denoising process

diffusion models

generative models

latent space

watermarking

AI generated summary

Embedding Watermarks in Stable Diffusion Models

This paper proposes a plug-and-play framework to embed watermarks in Stable Diffusion models without retraining. The watermarks are embedded in latent space and adapt to the denoising process. Results show effective balance of image quality and watermark invisibility, robustness to attacks, and generalization across SD versions.