Beyond the Safety Tax: Mitigating Unsafe Text-to-Image Generation via External Safety Rectification

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.

Key Contributions

SafePatch: a structurally isolated external safety module (cloned encoder architecture) that rectifies T2I generation without modifying base model weights, eliminating the 'Safety Tax' of concept entanglement
Aligned Counterfactual Safety (ACS) dataset: strictly paired unsafe/safe image pairs with identical benign semantics, enabling interpretable differential supervision for safety-specific rectification
Demonstrates 7% unsafe rate on I2P benchmark — substantially below baselines (~20%) — while maintaining image quality and semantic alignment under adversarial prompt attacks