Research map / Image generation / Text-to-image diffusion
JANUS: A Lightweight Framework for Jailbreaking Text-to-Image Models via Distribution Optimization
Haolun Zheng, Yu He, Tailun Chen, Shuo Shao, Zhixuan Chu, Hongbin Zhou, Lan Tao, Zhan Qin, Kui Ren
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 dual-distribution interference instead of a single distribution (ablation) | JANUS · SD3.5LT | ASR-8 (%): 26.87% -> 44.50% | Improved |
| use dual-distribution interference instead of a single distribution (ablation) | JANUS · DALL-E3 | ASR-8 (%): 0.69% -> 1.39% | Improved |
| use dual-distribution interference instead of a single distribution (ablation) | JANUS · SD3.5LT | TASR: 97.00% -> 94.25% | Hurt |
| use dynamic harmfulness-based reward instead of fixed reward (ablation) | JANUS · SD3.5LT | NSFW score: 0.286 -> 0.329 | Improved |
| use dynamic harmfulness-based reward instead of fixed reward (ablation) | JANUS · SD3.5LT | ASR-8 (%): 32.33% -> 44.50% | Improved |
| use dynamic harmfulness-based reward instead of fixed reward (ablation) | JANUS · DALL-E3 | NSFW score: 0.071 -> 0.082 | Improved |
| learn dynamic mixing alpha with RL instead of fixed alpha (ablation) | JANUS · SD3.5LT and DALL-E3 | ASR and NSFW score | Improved |
| use high fixed alpha (1.0) toward harmful distribution (ablation) | JANUS · SD3.5LT and DALL-E3 | TASR | Hurt |
| use low fixed alpha toward clean distribution (ablation) | JANUS · SD3.5LT and DALL-E3 | NSFW score | Hurt |
| use JANUS instead of PGJ on SD3.5LT | JANUS · SD3.5LT | ASR-8 (%): 17.15 -> 43.15 | Improved |
| use JANUS instead of QFA on DALL-E3 | JANUS · DALL-E3 | ASR-8 (%): 1.53 -> 3.39 | Improved |
| use JANUS instead of QFA on SDXL | JANUS · SDXL | ASR-8 (%): 24.26 -> 58.20 | Improved |
Closest papers by meaning
- T2V-OptJail: Discrete Prompt Optimization for Text-to-Video Jailbreak Attacks (NeurIPS 2025 · similarity 0.91)
- Perception-Guided Jailbreak Against Text-to-Image Models (AAAI 2025 · similarity 0.89)
- ColJailBreak: Collaborative Generation and Editing for Jailbreaking Text-to-Image Deep Generation (NeurIPS 2024 · similarity 0.88)
- MIDAS: Multi-Image Dispersion and Semantic Reconstruction for Jailbreaking MLLMs (ICLR 2026 · similarity 0.88)
- Distraction is All You Need for Multimodal Large Language Model Jailbreaking (CVPR 2025 · similarity 0.88)
- JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (NeurIPS 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.
