Research map / Image generation / Text-to-image diffusion
Value-Aligned Prompt Moderation via Zero-Shot Agentic Rewriting for Safe Image Generation
Xin Zhao, Xiaojun Chen, Bingshan Liu, Zeyao Liu, Zhendong Zhao, Xiaoyan Gu
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 VALOR prompt rewriting in front of the T2I model | SDV1.4 · I2P-Sexual | SAFE rate: 67.9% -> 92.7% | Improved |
| add VALOR prompt rewriting in front of the T2I model | SDV1.4 · 4chan | SAFE rate: 83.2% -> 99.6% | Improved |
| add word-level keyword detection (ablation) | VALOR · Value-412 | ACC: 88.8% -> 90.5% | Improved |
| add semantic-level detection (ablation) | VALOR · Value-412 | FNR: 16.6% -> 11.2% | Improved |
| add value-level detection (ablation) | VALOR · Value-412 | ACC: 88.8% -> 90.5% | Improved |
| add detection gating before rewriting instead of rewriting everything (ablation) | VALOR · Value-412 | FPR: 100.0% -> 6.5% | Improved |
| use a small rewriter (Qwen1.5-1.8B-Chat) instead of larger LLMs | VALOR · Intent-452 | FNR: up to 33.3% | Hurt |
| add VALOR rewriting on benign prompts | SDV1.4 · COCO (benign) | FID: 17.04 -> 15.89 | Improved |
| add LLM rewriting step to the pipeline | VALOR · Average time per prompt | latency overhead: 0.575 s to 2.54 s per prompt | Hurt |
| use VALOR instead of baseline defenses on jailbreak datasets | SDV1.4 · Jailbreak datasets (SneakyPrompt, QF-Attack, MMP-Attack, MMA-Diffusion) | NRR: 98.0% to 100.0% versus lowest baseline 12.7% | Improved |
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.
- SAFREE — SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation (ICLR 2025) · 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
- Focus-N-Fix: Region-Aware Fine-Tuning for Text-to-Image Generation (CVPR 2025 · similarity 0.88)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.87)
- Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts (ICML 2024 · similarity 0.87)
- ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time (ICLR 2025 · similarity 0.87)
- Safe + Safe = Unsafe? Exploring How Safe Images Can Be Exploited to Jailbreak Large Vision-Language Models (NeurIPS 2025 · similarity 0.87)
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.86)
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.
