Broken Memories: Detecting and Mitigating Memorization in Diffusion Models with Degraded Generations

2026-05-21

AI Summary

Diffusion models, while capable of generating high-quality images, are prone to memorizing training data, raising significant privacy and copyright concerns. This work addresses this issue by introducing numerical stability analysis into the study of memorization in diffusion models and proposes a novel paradigm for real-time memory detection and suppression that requires no modification to prompts or guidance. The method constructs an empirical stability region based on the norm of latent-space update steps and integrates an online detection algorithm with an adaptive de-memorization mechanism. Evaluated on Stable Diffusion 1.4, the approach achieves exceptional memory detection performance with an AUC exceeding 0.999 and reduces the post-mitigation memorization rate to 0%, imposing only a minimal computational overhead of approximately 0.01 seconds per image.

Abstract

While diffusion models excel at generating high-quality images, their tendency to memorize training data poses significant privacy and copyright risks. In this work, we for the first time identify that memorization induces internal numerical instability, often manifesting as visually ``broken'' artifacts. Inspired by stability analysis in numerical methods, we introduce empirical stability regions based on latent update norms to quantitatively characterize stable behavior during generation. Leveraging this, we propose a principled, on-the-fly framework for step-wise detection and adaptive mitigation. Our approach suppresses memorization without altering prompts or guidance, thereby preserving semantic fidelity and image quality. Extensive experiments on Stable Diffusion 1.4 demonstrate that our method achieves an AUC $>0.999$ detection performance and a $0.0\%$ memorization rate after mitigation with negligible overhead ($\approx0.01$s per image).

Problem

• memorization
• diffusion models
• privacy
• copyright
• data memorization

Innovation

• memorization detection
• diffusion models
• numerical stability
• on-the-fly mitigation
• latent update norms