Multimodal image editing systems miss risks in combined text and images

Text-Aware, Vision-Blind: How Multimodal Image Editing Systems Handle Unsafe Inputs

Cryptography and SecurityComputer Vision and Pattern RecognitionMachine Learning

Summary

Many tools that edit images using instructions don’t check both the picture and the text for unsafe content together. The authors created a test called EditSafetyBench to see how well these tools handle risky images and unsafe editing directions. They found that some systems refuse to make unsafe edits based on the text but often miss risks in the images. Different systems handle types of risky content very differently, especially for political and sexual content. This shows that better safety checks need to look at both pictures and text at the same time.

What this means in practice

Authors

Ziqing Yang, Yixin Wu, Michael Backes, Yang Zhang

Abstract

Multimodal image generation systems have made rapid progress and increasingly emphasize content safety. However, their safety mechanisms tend to rely on a single modality and lack joint consideration of both visual content and text prompts, particularly in instruction-guided image editing, where the uploaded image may itself carry risk. We thus introduce EditSafetyBench, a benchmark that measures system-level safety awareness of advanced image editing systems, containing 5,940 image-instruction pairs, covering safe and unsafe images across 11 risk categories and three edit intents. We then probe the benchmark dataset on four systems, including Gemini, GPT, Qwen API, and Qwen Local. Based on our evaluation, we observe a systematic text-vision safety mismatch: Gemini and GPT refuse 87.1\% and 81.6\% cases containing unsafe-intent editing prompts, while only refusing 24.2\% and 22.4\% cases with both unsafe images and neutral instructions. At the category level, most risks are considered inconsistently by different systems, e.g., GPT allows 70.4\% editing attempts of unsafe political content while Gemini only allows 34.8\%. Notably, sexual content shows the strongest visual sensitivity, likely benefiting from mature NSFW identification works. Our findings highlight the need for safety guardrails that jointly reason over both visual and textual modalities in image editing systems.