Silent Branding Attack: Trigger-free Data Poisoning Attack on Text-to-Image Diffusion Models
Sangwon Jang, June Suk Choi, Jaehyeong Jo, Kimin Lee, Sung Ju Hwang
2025-03-14
Summary
This paper talks about a sneaky way to trick AI image generators into adding brand logos or symbols to pictures without being told to, by secretly altering their training data.
What's the problem?
AI image tools are vulnerable to attacks where hidden changes in training data make them generate unwanted content, like logos, even when users don’t ask for them.
What's the solution?
The 'Silent Branding Attack' automatically hides logos in training images so well that AI learns to add them naturally, without users mentioning them in prompts.
Why it matters?
This raises concerns about AI misuse for hidden advertising or spreading harmful symbols, forcing developers to secure models against such tricks.
Abstract
Text-to-image diffusion models have achieved remarkable success in generating high-quality contents from text prompts. However, their reliance on publicly available data and the growing trend of data sharing for fine-tuning make these models particularly vulnerable to data poisoning attacks. In this work, we introduce the Silent Branding Attack, a novel data poisoning method that manipulates text-to-image diffusion models to generate images containing specific brand logos or symbols without any text triggers. We find that when certain visual patterns are repeatedly in the training data, the model learns to reproduce them naturally in its outputs, even without prompt mentions. Leveraging this, we develop an automated data poisoning algorithm that unobtrusively injects logos into original images, ensuring they blend naturally and remain undetected. Models trained on this poisoned dataset generate images containing logos without degrading image quality or text alignment. We experimentally validate our silent branding attack across two realistic settings on large-scale high-quality image datasets and style personalization datasets, achieving high success rates even without a specific text trigger. Human evaluation and quantitative metrics including logo detection show that our method can stealthily embed logos.