You Don't Need All That Attention: Surgical Memorization Mitigation in Text-to-Image Diffusion Models

Researchers introduce GUARD, a novel framework to prevent text-to-image AI models from memorizing and reproducing training data that could lead to privacy or copyright issues. The method uses attention attenuation to guide image generation away from original training data while maintaining prompt alignment and image quality.

• →GUARD framework mitigates memorization in text-to-image diffusion models without compromising generation quality
• →The system uses cross-attention attenuation to automatically identify and adjust problematic prompt positions
• →Method operates at inference-time with surgical precision, working dynamically per-prompt
• →GUARD demonstrates state-of-the-art results across multiple architectures for both verbatim and template memorization
• →The approach addresses growing privacy and copyright concerns in AI-generated content