ForceForget: Reinforcement Concept Removal for Enhancing Safety in Text-to-Image Models

With the advance of generative AI, the text-to-image (T2I) model has the ability to generate various contents. However, T2I models still can generate unsafe contents. To alleviate this issue, various concept erasing methods are proposed. However, existing methods tend to excessively erase unsafe concepts and suppress benign concepts contained in harmful prompts, which can negatively affect model utility. In this paper, we focus on eliminating unsafe content while maintaining model capability in safe semantic meaning interpretation by optimizing the concept erasing reward (CER) with reinforcement learning. To avoid overly content erasure, we introduce the Safe Adapter to project partial text embedding for efficient concept regulation in cross-attention layers. Extensive experiments conducted on different datasets demonstrate the effectiveness of the proposed method in alleviating unsafe content generation while preserving the high fidelity of benign images compared with existing state-of-the-art (SOTA) concept erasing methods. In terms of robustness, our method outperforms counterparts against red-teaming tools. Moreover, we showcase the proposed approach is more effective in emerging image-to-image (I2I) scenarios compared with others. Lastly, we extend our method to erase general concepts, such as artistic styles and objects. Disclaimer: This paper includes discussions of sexually explicit content that may be offensive to certain readers. All images used in this work are synthesized or from public datasets.

From input texts, generative AI systems can produce realistic images. However, there are concerns about misuse since these systems can potentially produce harmful content. Reducing the production of hazardous content while preserving usability for normal applications is essential. In this work, we developed a new training approach that helps image generation models better distinguish between unsafe and safe meanings in text prompts. Instead of broadly suppressing all sensitive concepts, our method carefully erases only the harmful parts. We achieve this by combining reinforcement learning with a lightweight “safe adapter” that selectively adjusts how the model interprets input text. Experiments show that our method reduces unsafe image generation effectively while still producing high-quality benign images. Apart from ensuring safety in text-to-image generation, our method also works in image-to-image editing. Overall, this research helps make generative AI systems safer without unnecessarily reducing their capabilities.