Research map / Image generation / Text-to-image diffusion
Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply SLD Hyp-Strong safety guidance | Stable Diffusion 1.4 · I2P overall | inappropriate probability: 0.39 -> 0.13 | Helped |
| Apply SLD Hyp-Max safety guidance | Stable Diffusion 1.4 · I2P overall | inappropriate probability: 0.39 -> 0.09 | Helped |
| Increase safety-guidance strength from weak to maximum (ablation) | Stable Diffusion 1.4 · I2P overall | inappropriate probability: 0.29 -> 0.09 | Helped |
| Use the safety concept as a negative prompt instead of SLD | Stable Diffusion 1.4 · I2P overall | inappropriate probability: 0.13 -> 0.18 | Hurt |
| Apply SLD Hyp-Strong safety guidance | Stable Diffusion 2.0 · I2P overall | inappropriate probability: 0.40 -> 0.13 | Helped |
| Combine dataset filtering with SLD | Stable Diffusion 2.0 · I2P overall | inappropriate probability: 0.40 -> 0.10 | Helped |
| Apply SLD Hyp-Strong safety guidance | Stable Diffusion 1.4 · Body-Ethnicity | nude image probability: 0.36 -> 0.09 | Helped |
| Apply SLD Hyp-Strong safety guidance | Stable Diffusion 1.4 · COCO | FID-30k: 14.43 -> 18.28 | Hurt |
| Apply SLD safety guidance | Stable Diffusion 1.4 · DrawBench | image fidelity user preference: 63.13% | Helped |
| Apply SLD safety guidance | Stable Diffusion 1.4 · DrawBench | text alignment user preference: 59.62% | Helped |
| Increase the warm-up period (ablation) | SLD · Qualitative image editing | image composition preservation | Helped |
| Increase the safety threshold (ablation) | SLD · Qualitative image editing | image change magnitude | Helped |
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 0.91)
- Growth Inhibitors for Suppressing Inappropriate Image Concepts in Diffusion Models (ICLR 2025 · similarity 0.90)
- Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts (ICML 2024 · similarity 0.90)
- Towards Resilient Safety-driven Unlearning for Diffusion Models against Downstream Fine-tuning (NeurIPS 2025 · similarity 0.89)
- Six-CD: Benchmarking Concept Removals for Text-to-image Diffusion Models (CVPR 2025 · similarity 0.89)
- Leveraging Catastrophic Forgetting to Develop Safe Diffusion Models against Malicious Finetuning (NeurIPS 2024 · 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.