Beyond the Safety Tax: Mitigating Unsafe Text-to-Image Generation via External Safety Rectification
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917208866684928 |
|---|---|
| author | Meng, Xiangtao Dong, Yingkai Yu, Ning Wang, Li Li, Zheng Guo, Shanqing |
| author_facet | Meng, Xiangtao Dong, Yingkai Yu, Ning Wang, Li Li, Zheng Guo, Shanqing |
| contents | Text-to-image (T2I) generative models have achieved remarkable visual fidelity, yet remain vulnerable to generating unsafe content. Existing safety defenses typically intervene internally within the generative model, but suffer from severe concept entanglement, leading to degradation of benign generation quality, a trade-off we term the Safety Tax. To overcome this limitation, we advocate a paradigm shift from destructive internal editing to external safety rectification. Following this principle, we propose SafePatch, a structurally isolated safety module that performs external, interpretable rectification without modifying the base model. The core backbone of SafePatch is architecturally instantiated as a trainable clone of the base model's encoder, allowing it to inherit rich semantic priors and maintain representation consistency. To enable interpretable safety rectification, we construct a strictly aligned counterfactual safety dataset (ACS) for differential supervision training. Across nudity and multi-category benchmarks and recent adversarial prompt attacks, SafePatch achieves robust unsafe suppression (7% unsafe on I2P) while preserving image quality and semantic alignment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_21099 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Beyond the Safety Tax: Mitigating Unsafe Text-to-Image Generation via External Safety Rectification Meng, Xiangtao Dong, Yingkai Yu, Ning Wang, Li Li, Zheng Guo, Shanqing Computer Vision and Pattern Recognition Artificial Intelligence Cryptography and Security Text-to-image (T2I) generative models have achieved remarkable visual fidelity, yet remain vulnerable to generating unsafe content. Existing safety defenses typically intervene internally within the generative model, but suffer from severe concept entanglement, leading to degradation of benign generation quality, a trade-off we term the Safety Tax. To overcome this limitation, we advocate a paradigm shift from destructive internal editing to external safety rectification. Following this principle, we propose SafePatch, a structurally isolated safety module that performs external, interpretable rectification without modifying the base model. The core backbone of SafePatch is architecturally instantiated as a trainable clone of the base model's encoder, allowing it to inherit rich semantic priors and maintain representation consistency. To enable interpretable safety rectification, we construct a strictly aligned counterfactual safety dataset (ACS) for differential supervision training. Across nudity and multi-category benchmarks and recent adversarial prompt attacks, SafePatch achieves robust unsafe suppression (7% unsafe on I2P) while preserving image quality and semantic alignment. |
| title | Beyond the Safety Tax: Mitigating Unsafe Text-to-Image Generation via External Safety Rectification |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Cryptography and Security |
| url | https://arxiv.org/abs/2508.21099 |