Comprehensive Assessment and Analysis for NSFW Content Erasure in Text-to-Image Diffusion Models

2025-02-18

AI Summary

Text-to-image diffusion models frequently generate NSFW content, posing significant risks to safe deployment; however, existing concept-erasure methods lack systematic evaluation. This work presents the first comprehensive empirical study of NSFW and its fine-grained subcategories’ erasure, evaluating 11 mainstream techniques and 14 variants across six dimensions: erasure rate, image fidelity, semantic alignment, over-erasure, and robustness against both explicit and implicit unsafe prompts. We propose a novel multidimensional evaluation framework incorporating implicit prompt sensitivity analysis, classifier-based toxicity assessment, and a fine-grained NSFW subcategory benchmark; we also release the first open-source, reproducible evaluation toolkit. Experiments reveal that current methods consistently fail under implicit unsafe prompts and frequently induce over-erasure. Our study substantially advances the credibility and practical safety of NSFW erasure techniques for real-world deployment.

Technology Category

• Natural Language Processing: Safety and Robustness
• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Text-to-image (T2I) diffusion models have gained widespread application across various domains, demonstrating remarkable creative potential. However, the strong generalization capabilities of these models can inadvertently led they to generate NSFW content even with efforts on filtering NSFW content from the training dataset, posing risks to their safe deployment. While several concept erasure methods have been proposed to mitigate this issue, a comprehensive evaluation of their effectiveness remains absent. To bridge this gap, we present the first systematic investigation of concept erasure methods for NSFW content and its sub-themes in text-to-image diffusion models. At the task level, we provide a holistic evaluation of 11 state-of-the-art baseline methods with 14 variants. Specifically, we analyze these methods from six distinct assessment perspectives, including three conventional perspectives, i.e., erasure proportion, image quality, and semantic alignment, and three new perspectives, i.e., excessive erasure, the impact of explicit and implicit unsafe prompts, and robustness. At the tool level, we perform a detailed toxicity analysis of NSFW datasets and compare the performance of different NSFW classifiers, offering deeper insights into their performance alongside a compilation of comprehensive evaluation metrics. Our benchmark not only systematically evaluates concept erasure methods, but also delves into the underlying factors influencing their performance at the insight level. By synthesizing insights from various evaluation perspectives, we provide a deeper understanding of the challenges and opportunities in the field, offering actionable guidance and inspiration for advancing research and practical applications in concept erasure.

Problem

• Evaluates NSFW content erasure in text-to-image models
• Assesses 11 methods across six evaluation perspectives
• Analyzes toxicity and classifier performance in NSFW datasets

Innovation

• Systematically evaluates concept erasure methods
• Analyzes NSFW datasets for toxicity
• Compares NSFW classifiers' performance