Research map / Image generation / Text-to-image diffusion
Responsible Text-to-Image Diffusion: Interpretable and Linearly Controllable Semantics for Fair and Safe Generation
Sayedmoslem Shokrolahi, Jae-Mo Kang, Il-Min Kim
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 learned gender concept vectors with equal probability | PixArt-alpha · WinoBias gender | deviation ratio: 0.82 -> 0.05 | Improved |
| add learned concept vectors to FLUX | FLUX.1-dev · WinoBias race | deviation ratio: 0.84 -> 0.08 | Improved |
| add concept vectors for gender and race jointly | PixArt-alpha · WinoBias gender x race | deviation ratio: 0.73 -> 0.02 | Improved |
| add anti-unsafe concept vectors | SDXL · I2P unsafe images | percentage unsafe: 56 -> 14 | Improved |
| add anti-unsafe concept vectors | FLUX.1-dev · I2P unsafe images | percentage unsafe: 64 -> 20 | Improved |
| use concept-alignment loss instead of simple reconstruction loss (ablation) | concept vector learning · spurious attribute 'old' leakage | CLIP delta old: 1.02 -> 0.61 | Improved |
| use concept-alignment loss instead of simple reconstruction loss (ablation) | concept vector learning · target attribute 'female' | CLIP delta female: 1.28 -> 1.45 | Improved |
| restrict intervention to Top-5 heads instead of all heads in selected layers (ablation) | PixArt-alpha · WinoBias | CLIP score: 31.62 -> 34.01 | Improved |
| extend intervention to all layers instead of selected layers (ablation) | PixArt-alpha · WinoBias | CLIP score: 31.62 -> 29.61 | Hurt |
| keep Top-5 heads in the intervention instead of excluding them (ablation) | PixArt-alpha · WinoBias | deviation ratio: 0.02 -> 0.01 | Improved |
| increase training images beyond about 3,000 (1,500 for SDXL) (ablation) | concept vector learning · doctor prompt | deviation ratio | No clear effect |
| vary the target-removed prompt template (five templates) (ablation) | concept vector learning · a photo of a doctor | deviation ratio | No clear effect |
Closest papers by meaning
- The Hidden Language of Diffusion Models (ICLR 2024 · similarity 0.91)
- Compositional Text-to-Image Synthesis with Attention Map Control of Diffusion Models (AAAI 2024 · similarity 0.91)
- Rare Text Semantics Were Always There in Your Diffusion Transformer (NeurIPS 2025 · similarity 0.91)
- Plug-and-Play Interpretable Responsible Text-to-Image Generation via Dual-Space Multi-facet Concept Control (CVPR 2025 · similarity 0.91)
- T2I-Adapter: Learning Adapters to Dig Out More Controllable Ability for Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.90)
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 2024 · similarity 0.90)
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.
