M-ErasureBench: A Comprehensive Multimodal Evaluation Benchmark for Concept Erasure in Diffusion Models

TL;DR

This paper introduces M-ErasureBench, a comprehensive benchmark for evaluating concept erasure in diffusion models across multiple input modalities, revealing vulnerabilities and proposing IRECE to enhance robustness.

Contribution

It presents the first multimodal evaluation framework for concept erasure in diffusion models and introduces IRECE, a method to improve erasure robustness across different input types.

Findings

Existing methods perform well on text prompts but poorly on learned embeddings and latents.

Concept Reproduction Rate exceeds 90% in white-box scenarios without robust defenses.

IRECE reduces CRR by up to 40% while maintaining visual quality.

Abstract

Text-to-image diffusion models may generate harmful or copyrighted content, motivating research on concept erasure. However, existing approaches primarily focus on erasing concepts from text prompts, overlooking other input modalities that are increasingly critical in real-world applications such as image editing and personalized generation. These modalities can become attack surfaces, where erased concepts re-emerge despite defenses. To bridge this gap, we introduce M-ErasureBench, a novel multimodal evaluation framework that systematically benchmarks concept erasure methods across three input modalities: text prompts, learned embeddings, and inverted latents. For the latter two, we evaluate both white-box and black-box access, yielding five evaluation scenarios. Our analysis shows that existing methods achieve strong erasure performance against text prompts but largely fail under learned embeddings and inverted latents, with Concept Reproduction Rate (CRR) exceeding 90% in the white-box setting. To address these vulnerabilities, we propose IRECE (Inference-time Robustness Enhancement for Concept Erasure), a plug-and-play module that localizes target concepts via cross-attention and perturbs the associated latents during denoising. Experiments demonstrate that IRECE consistently restores robustness, reducing CRR by up to 40% under the most challenging white-box latent inversion scenario, while preserving visual quality. To the best of our knowledge, M-ErasureBench provides the first comprehensive benchmark of concept erasure beyond text prompts. Together with IRECE, our benchmark offers practical safeguards for building more reliable protective generative models.