# SafetyPairs: Isolating Safety Critical Image Features with Counterfactual Image Generation

arXiv: https://arxiv.org/abs/2510.21120 · HTML: https://mlsanity.com/paper/2510.21120/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 26 of its 34 references that are in the map.
- First posted on arXiv 24 Oct 2025.

## Builds on (most cited first, 26 of 26)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [DINOv2: Learning Robust Visual Features without Supervision](https://mlsanity.com/paper/2304.07193/) (2023)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [Sigmoid Loss for Language Image Pre-Training](https://mlsanity.com/paper/2303.15343/) (2023)
- [InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models](https://mlsanity.com/paper/2504.10479/) (2025)
- [FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation](https://mlsanity.com/paper/2305.14251/) (2023)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space](https://mlsanity.com/paper/2506.15742/) (2025)
- [Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMs](https://mlsanity.com/paper/2401.06209/) (2024)
- [The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes](https://mlsanity.com/paper/2005.04790/) (2020)
- [MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models](https://mlsanity.com/paper/2311.17600/) (2023)
- [LLM Self Defense: By Self Examination, LLMs Know They Are Being Tricked](https://mlsanity.com/paper/2308.07308/) (2023)
- [Deepfake Generation and Detection: A Benchmark and Survey](https://mlsanity.com/paper/2403.17881/) (2024)
- [SafeGen: Mitigating Sexually Explicit Content Generation in Text-to-Image Models](https://mlsanity.com/paper/2404.06666/) (2024)
- [UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Images](https://mlsanity.com/paper/2405.03486/) (2024)
- [LlavaGuard: An Open VLM-based Framework for Safeguarding Vision Datasets and Models](https://mlsanity.com/paper/2406.05113/) (2024)
- [LANCE: Stress-testing Visual Models by Generating Language-guided Counterfactual Images](https://mlsanity.com/paper/2305.19164/) (2023)
- [Navigating the Safety Landscape: Measuring Risks in Finetuning Large Language Models](https://mlsanity.com/paper/2405.17374/) (2024)
- [Training Unbiased Diffusion Models From Biased Dataset](https://mlsanity.com/paper/2403.01189/) (2024)
- [Robust CLIP-Based Detector for Exposing Diffusion Model-Generated Images](https://mlsanity.com/paper/2404.12908/) (2024)
- [Generative AI Misuse: A Taxonomy of Tactics and Insights from Real-World Data](https://mlsanity.com/paper/2406.13843/) (2024)
- [AlignGuard: Scalable Safety Alignment for Text-to-Image Generation](https://mlsanity.com/paper/2412.10493/) (2024)
- [MSTS: A Multimodal Safety Test Suite for Vision-Language Models](https://mlsanity.com/paper/2501.10057/) (2025)

## Built on by (most cited first, 2 of 2)

- [Cross-modal Counterfactual Explanations: Uncovering Decision Factors and Dataset Biases in Subjective Classification](https://mlsanity.com/paper/2512.18864/) (2025)
- [Enabling Preference-driven Unlearning in Few-step Distilled Text-to-Image Diffusion Models](https://mlsanity.com/paper/2610.10859/) (2026)

## Abstract (arXiv)

What exactly makes a particular image unsafe? Systematically differentiating between benign and problematic images is a challenging problem, as subtle changes to an image, such as an insulting gesture or symbol, can drastically alter its safety implications. However, existing image safety datasets are coarse and ambiguous, offering only broad safety labels without isolating the specific features that drive these differences. We introduce SafetyPairs, a scalable framework for generating counterfactual pairs of images, that differ only in the features relevant to the given safety policy, thus flipping their safety label. By leveraging image editing models, we make targeted changes to images that alter their safety labels while leaving safety-irrelevant details unchanged. Using SafetyPairs, we construct a new safety benchmark, which serves as a powerful source of evaluation data that highlights weaknesses in vision-language models' abilities to distinguish between subtly different images. Beyond evaluation, we find our pipeline serves as an effective data augmentation strategy that improves the sample efficiency of training lightweight guard models. We release a benchmark containing over 3,020 SafetyPair images spanning a diverse taxonomy of 9 safety categories, providing the first systematic resource for studying fine-grained image safety distinctions.