Research map / Adversarial robustness and security / Attacks on text and graphs
AutoPrompt: Automated Red-Teaming of Text-to-Image Models via LLM-Driven Adversarial Prompts
Yufan Liu, Wanqian Zhang, Huashan Chen, Lin Wang, Xiaojun Jia, Zheng Lin, Weiping Wang
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 unsafe image alignment (ablation) | AutoPrompt · I2P nudity (ESD) | RSR: 38.5% -> 61.5% | Improved |
| add unsafe word alignment (ablation) | AutoPrompt · I2P nudity (ESD) | RSR: 30.5% -> 61.5% | Improved |
| add perplexity constraint (ablation) | AutoPrompt · I2P nudity (ESD) | RSR: 35% -> 61.5% | Improved |
| add perplexity constraint (ablation) | AutoPrompt · I2P nudity (ESD) | PPL avg: 0.198 -> 0.167 | Improved |
| add banned-token penalty (ablation) | AutoPrompt · I2P nudity (ESD) | RSR: 9.5% -> 61.5% | Improved |
| add banned-token penalty (ablation) | AutoPrompt · I2P nudity (ESD) | blocking rate: 87% -> 2% | Improved |
| add prior suffix (ablation) | AutoPrompt · I2P nudity (SLD-MAX) | RSR: 18.5% -> 70.5% | Improved |
| add prior suffix (ablation) | AutoPrompt · I2P nudity (ESD) | RSR: 17% -> 61.50% | Improved |
| add alternating fine-tuning of suffix generator over autoregressive decoding (ablation) | AutoPrompt with prior suffix · I2P nudity (SLD-MAX) | RSR: 39% -> 70.50% | Improved |
| add alternating fine-tuning over prior suffix alone (ablation) | prior suffix · I2P nudity (SLD-MAX) | RSR: 6.5% -> 70.50% | Improved |
| use APT instead of P4D-Union | red-teaming method · I2P nudity (ESD) | RSR: 41.50% -> 61.50% | Improved |
| use APT instead of P4D-Union | red-teaming method · I2P violence (SLD-MAX) | RSR: 30.99% -> 20.66% | Hurt |
Closest papers by meaning
- AdvPrompter: Fast Adaptive Adversarial Prompting for LLMs (ICML 2025 · similarity 0.91)
- ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users (NeurIPS 2024 · similarity 0.88)
- Red-Teaming Text-to-Image Systems by Rule-based Preference Modeling (NeurIPS 2025 · similarity 0.88)
- An LLM can Fool Itself: A Prompt-Based Adversarial Attack (ICLR 2024 · similarity 0.88)
- RISE: Red-teaming via Iterative Strategy Evolution for Modern Text-to-Image Models (NeurIPS 2026 · similarity 0.88)
- One Prompt Word is Enough to Boost Adversarial Robustness for Pre-trained Vision-Language Models (CVPR 2024 · 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.
