Research map / Image generation / Text-to-image diffusion
FairGen: Enhancing Fairness in Text-to-Image Diffusion Models via Self-Discovering Latent Directions
Yilei Jiang, Wei-Hong Li, Yiyuan Zhang, Minghong Cai, Xiangyu Yue
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 |
|---|---|---|---|
| restrict adapters to cross-attention layers instead of all layers (ablation) | FairGen · Intersectional debiasing | FD: 0.165 -> 0.047 | Improved |
| restrict adapters to cross-attention layers instead of non-CA layers (ablation) | FairGen · Intersectional debiasing | FD: 0.158 -> 0.047 | Improved |
| restrict adapters to cross-attention layers instead of all layers (ablation) | FairGen · Intersectional debiasing | BRISQUE: 35.00 -> 39.50 | Improved |
| add orthogonal regularization (ablation) | FairGen · Intersectional debiasing | FD: 0.143 -> 0.047 | Improved |
| add orthogonal regularization (ablation) | FairGen · Intersectional debiasing | CLIPsim: 0.29 -> 0.36 | Improved |
| add orthogonal regularization (ablation) | FairGen · Intersectional debiasing | BRISQUE: 37.92 -> 39.50 | Improved |
| FairGen vs Original SD | Original SD · Gender debiasing (0.5,0.5) | FD: 0.424 -> 0.003 | Improved |
| FairGen vs Original SD | Original SD · Gender debiasing (0.2,0.8) | FD: 0.847 -> 0.005 | Improved |
| FairGen vs Original SD | Original SD · Racial debiasing (0.25 each) | FD: 0.384 -> 0.095 | Improved |
| FairGen vs Original SD | Original SD · Racial debiasing (0.4,0.3,0.2,0.1) | FD: 0.497 -> 0.150 | Improved |
| FairGen vs Original SD | Original SD · Intersectional debiasing | FD: 0.214 -> 0.047 | Improved |
| train adapters on SD v2.1, test on v1.4 | FairGen · Gender debiasing transfer | FD: 0.003 -> 0.006 | No clear effect |
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.
- ITI-GEN — ITI-GEN: Inclusive Text-to-Image Generation (ICCV 2023) · better on 5 of 5 shared results
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
- Balancing Act: Distribution-Guided Debiasing in Diffusion Models (CVPR 2024 · similarity 0.93)
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.92)
- Fair Text-to-Image Diffusion via Fair Mapping (AAAI 2025 · similarity 0.90)
- Finetuning Text-to-Image Diffusion Models for Fairness (ICLR 2024 · similarity 0.90)
- LightFair: Towards an Efficient Alternative for Fair T2I Diffusion via Debiasing Pre-trained Text Encoders (NeurIPS 2025 · similarity 0.90)
- Rethinking Training for De-biasing Text-to-Image Generation: Unlocking the Potential of Stable Diffusion (CVPR 2025 · 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.
