Research map / Image generation / Text-to-image diffusion
T2I-RiskyPrompt: A Benchmark for Safety Evaluation, Attack, and Defense on Text-to-Image Model
Chenyu Zhang, Tairen Zhang, Lanjun Wang, Ruidong Chen, Wenhui Li, Anan 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 |
|---|---|---|---|
| use a more capable T2I model | T2I models · T2I-RiskyPrompt | risk ratio: SD1.4 0.790 -> SD3 0.923 | Hurt |
| apply MACE fine-tuning defense | SD1.4 · T2I-RiskyPrompt pornography | risk ratio: 0.976 -> 0.190 | Improved |
| apply TRCE defense | SD1.4 · T2I-RiskyPrompt pornography | risk ratio: 0.976 -> 0.032 | Improved |
| apply SPEED model editing | SD1.4 · T2I-RiskyPrompt copyright | risk ratio: 0.826 -> 0.854 | No clear effect |
| apply SLD inference guidance | SD1.4 · T2I-RiskyPrompt pornography | risk ratio: 0.976 -> 0.874 | Mixed |
| apply stronger defense (MACE) | SD1.4 · T2I-RiskyPrompt copyright/political | CLIP-S: 30.97 -> 23.91 | Hurt |
| apply SPEED for political concepts | SD1.4 · T2I-RiskyPrompt political | risk ratio: 0.914 -> 0.201 | Improved |
| use keyword filter | SD1.4 · T2I-RiskyPrompt avg | risk ratio: 0.790 -> 0.119 | Improved |
| use NSFW-I image filter | SD1.4 · T2I-RiskyPrompt avg | risk ratio: 0.790 -> 0.662 | Improved |
| ensemble four filters | SD1.4 · T2I-RiskyPrompt avg | risk ratio: 0.119 -> 0.089 | Improved |
| apply jailbreak attacks (MMA, MJA) against filters | SD1.4 + filters · T2I-RiskyPrompt | risk ratio | Hurt |
Closest papers by meaning
- T2ISafety: Benchmark for Assessing Fairness, Toxicity, and Privacy in Image Generation (CVPR 2025 · similarity 0.93)
- T2VSafetyBench: Evaluating the Safety of Text-to-Video Generative Models (NeurIPS 2024 · similarity 0.91)
- OVERT: A Benchmark for Over-Refusal Evaluation on Text-to-Image Models (NeurIPS 2025 · similarity 0.90)
- Position: Towards Implicit Prompt For Text-To-Image Models (ICML 2024 · similarity 0.90)
- Six-CD: Benchmarking Concept Removals for Text-to-image Diffusion Models (CVPR 2025 · similarity 0.88)
- Video-SafetyBench: A Benchmark for Safety Evaluation of Video LVLMs (NeurIPS 2025 · similarity 0.88)
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.
