Research map / Image generation / Text-to-image diffusion
When Safety Collides: Resolving Multi-Category Harmful Conflicts in Text-to-Image Diffusion via Adaptive Safety Guidance
Yongli Xiang, Ziming Hong, Zhaoqing Wang, Xiangyu Zhao, Bo Han, Tongliang Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CASG dynamic category selection | SLD · I2P | harmful rate (%): 12.7 -> 10.2 | Improved |
| add CASG dynamic category selection | SLD · T2VSafetyBench | harmful rate (%): 25.2 -> 9.8 | Improved |
| add CASG dynamic category selection | SLD · CoProv2 | harmful rate (%): 7.1 -> 3.9 | Improved |
| add CASG category selection | SAFREE · I2P | harmful rate (%): 20.0 -> 18.9 | Improved |
| add CASG category selection | SAFREE · Unsafe Diffusion | harmful rate (%): 24.2 -> 17.5 | Improved |
| add CASG category selection | SAFREE · COCO | FID: 43.78 -> 46.25 | Hurt |
| add CASG category selection | SLD · COCO | FID: 52.11 -> 52.00 | No clear effect |
| use mismatched harmful keywords (hate) on sexual prompts (ablation) | SLD · sexual prompts | harmful rate (%): 67.2 -> 72.4 | Hurt |
| aggregate all harmful categories instead of matching category (sexual) (ablation) | SLD · sexual prompts | harmful rate (%): 3.2 -> 48.8 | Hurt |
| add hate to sexual keywords (ablation) | SLD · sexual prompts | harmful rate (%): 3.2 -> 5.8 | Hurt |
| replace dynamic CASG category with GPT-4o one-shot classification (ablation) | SLD · T2VSafetyBench | harmful rate (%): 9.8 -> 12.3 | Hurt |
| replace dynamic CASG category with QwenGuard classification (ablation) | SLD · Unsafe Diffusion | harmful rate (%): 9.8 -> 23.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAFREE — SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation (ICLR 2025) · better on 4 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Detect-and-Guide: Self-regulation of Diffusion Models for Safe Text-to-Image Generation via Guideline Token Optimization (CVPR 2025 · similarity 0.90)
- T2ISafety: Benchmark for Assessing Fairness, Toxicity, and Privacy in Image Generation (CVPR 2025 · similarity 0.89)
- OVERT: A Benchmark for Over-Refusal Evaluation on Text-to-Image Models (NeurIPS 2025 · similarity 0.88)
- Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts (ICML 2024 · similarity 0.88)
- Towards Resilient Safety-driven Unlearning for Diffusion Models against Downstream Fine-tuning (NeurIPS 2025 · similarity 0.87)
- Training-Free Safe Text Embedding Guidance for Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
