Research map / Image generation / Text-to-image diffusion
UniNDM: A Unified Noise-Driven Detection and Mitigation Framework Against Sexual Content in Text-to-Image Generation
Yao Huang, Yitong Sun, Huanran Chen, Ruochen Zhang, Shouwei Ruan, Ranjie Duan, Maoxun Yuan, Yinpeng Dong, Hui Xue, Xiaochun Cao, Xingxing Wei
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 initial noise optimization to fixed negative guidance (ablation) | UniNDM(U) on SDv1.4 · I2P | ASR: 33.1% -> 20.5% | Improved |
| replace fixed negative prompt with LLM-adaptive negative prompt (ablation) | UniNDM(U) on SDv1.4 · I2P | ASR: 33.1% -> 28.8% | Improved |
| add adaptive negative guidance and noise optimization together (full method) (ablation) | SDv1.4 · I2P | ASR: 60.7% -> 9.7% | Improved |
| add initial noise optimization alone (ablation) | SDv1.4 · I2P | ASR: 60.7% -> 31.2% | Improved |
| add adaptive negative prompt on top of noise optimization (full vs Neg+Noise) (ablation) | UniNDM(U) on SDv1.4 · I2P | ASR: 20.5% -> 9.7% | Improved |
| add initial noise optimization alone (ablation) | SDv3 · I2P | ASR: 37.2% -> 35.8% | No clear effect |
| add original (U-Net style) negative guidance to noise optimization (ablation) | SDv3 · I2P | ASR: 35.8% -> 13.7% | Improved |
| replace original negative guidance with global steering vector (ablation) | UniNDM(D) on SDv3 · I2P | ASR: 13.7% -> 11.8% | Improved |
| add region-constrained steering to global steering plus noise (ablation) | UniNDM(D) on SDv3 · I2P | ASR: 11.8% -> 9.9% | Improved |
| use LLM adaptive negative prompt instead of fixed 'nudity' (ablation) | UniNDM(U) on SDv1.4 · SP(N) | ASR: 27.5% -> 10.0% | Improved |
| use LLM adaptive negative prompt instead of fixed 'nudity' (ablation) | UniNDM(U) on SDv1.4 · SP(P) | ASR: 26.5% -> 11.0% | Improved |
| use first-step predicted noise detector instead of Qwen-3.5-standard | SDv1.4 · Detection avg | accuracy: 69.8% -> 95.1% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Safree — SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation (ICLR 2025) · better on 3 of 3 shared results
- Latent Guard — Latent Guard: a Safety Framework for Text-to-image Generation (ECCV 2024) · better on 1 of 1 shared result
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 0.92)
- Mitigating Sexual Content Generation via Embedding Distortion in Text-conditioned Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Growth Inhibitors for Suppressing Inappropriate Image Concepts in Diffusion Models (ICLR 2025 · similarity 0.88)
- Training-Free Safe Text Embedding Guidance for Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts (ICML 2024 · similarity 0.88)
- Adaptive Median Smoothing: Adversarial Defense for Unlearned Text-to-Image Diffusion Models at Inference Time (ICML 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.
