SemBind: Binding Diffusion Watermarks to Semantics Against Black-Box Forgery Attacks
Xin Zhang, Zijin Yang, Kejiang Chen, Linfeng Ma, Weiming Zhang, Nenghai Yu
TL;DR
SemBind addresses the vulnerability of latent-based diffusion watermarks to black-box forgery by binding watermark signals to image semantics through a learned semantic masker. The framework consists of a two-stage semantic masker training pipeline and a mask-based semantic binding operation that multiplicatively modulates the initial latent of a watermark, preserving image quality while enabling a tunable anti-forgery strength via a mask ratio. Empirically, SemBind reduces false acceptance under imprinting and reprompting attacks across four latent-based schemes, while maintaining FID and CLIP metrics and preserving provable undetectability when the underlying watermark is undetectable. The approach generalizes across different diffusion backbones and datasets, offering a practical, controllable defense for provenance and copyright authentication in real-world deployment.
Abstract
Latent-based watermarks, integrated into the generation process of latent diffusion models (LDMs), simplify detection and attribution of generated images. However, recent black-box forgery attacks, where an attacker needs at least one watermarked image and black-box access to the provider's model, can embed the provider's watermark into images not produced by the provider, posing outsized risk to provenance and trust. We propose SemBind, the first defense framework for latent-based watermarks that resists black-box forgery by binding latent signals to image semantics via a learned semantic masker. Trained with contrastive learning, the masker yields near-invariant codes for the same prompt and near-orthogonal codes across prompts; these codes are reshaped and permuted to modulate the target latent before any standard latent-based watermark. SemBind is generally compatible with existing latent-based watermarking schemes and keeps image quality essentially unchanged, while a simple mask-ratio parameter offers a tunable trade-off between anti-forgery strength and robustness. Across four mainstream latent-based watermark methods, our SemBind-enabled anti-forgery variants markedly reduce false acceptance under black-box forgery while providing a controllable robustness-security balance.