Research map / Image generation / Text-to-image diffusion
GenBreak: Red Teaming Text-to-Image Generation Using Large Language Models
Zilong Wang, Xiang Zheng, Xiaosen Wang, Bo Wang, Xingjun Ma
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 GenBreak (SFT + GRPO with multi-reward) instead of CRT | red-team LLM · Safeguarded SD 2.1 nudity | TBR (%): 2.9 -> 60.8 | Improved |
| use GenBreak instead of CRT | red-team LLM · Safeguarded SD 2.1 violence | TBR (%): 0.2 -> 89.7 | Improved |
| use GenBreak instead of CRT | red-team LLM · Safeguarded SD 2.1 hate | TBR (%): 2.9 -> 84.6 | Improved |
| use GenBreak instead of CRT | red-team LLM · Safeguarded SD 2.1 nudity | TCBR (%): 0.0 -> 57.9 | Improved |
| use GenBreak instead of CRT | red-team LLM · Safeguarded SD 2.1 nudity | lexical diversity: 0.863 -> 0.713 | Mixed |
| use GenBreak instead of vanilla RL | red-team LLM · Safeguarded SD 2.1 nudity | image toxicity: 0.842 -> 0.805 | Hurt |
| transfer GenBreak prompts to Leonardo.Ai versus CRT | red-team LLM · Leonardo.Ai nudity | TBR (%): 0 -> 70 | Improved |
| transfer GenBreak prompts to fal.ai hate versus CRT | red-team LLM · fal.ai hate | TBR (%): 90 -> 75 | Hurt |
| remove the bypass reward (ablation) | GenBreak RL · Safeguarded SD 2.1 nudity | bypass ability | Improved |
| add the clean reward (ablation) | GenBreak RL · Safeguarded SD 2.1 nudity | stealthiness | Improved |
| add lexical and semantic prompt diversity rewards (ablation) | GenBreak RL · Safeguarded SD 2.1 nudity | TCBR | Improved |
| add image diversity reward (ablation) | GenBreak RL · Safeguarded SD 2.1 nudity | image diversity | Mixed |
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.
- MMA-Diffusion — MMA-Diffusion: MultiModal Attack on Diffusion Models (CVPR 2024) · better on 3 of 3 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
- ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users (NeurIPS 2024 · similarity 0.93)
- T2ISafety: Benchmark for Assessing Fairness, Toxicity, and Privacy in Image Generation (CVPR 2025 · similarity 0.90)
- Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts (ICML 2024 · similarity 0.89)
- ColJailBreak: Collaborative Generation and Editing for Jailbreaking Text-to-Image Deep Generation (NeurIPS 2024 · similarity 0.89)
- T2VSafetyBench: Evaluating the Safety of Text-to-Video Generative Models (NeurIPS 2024 · similarity 0.89)
- RISE: Red-teaming via Iterative Strategy Evolution for Modern Text-to-Image Models (NeurIPS 2026 · similarity 0.89)
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.
