Research map / Image generation / Text-to-image diffusion
Stable Bias: Evaluating Societal Representations 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 |
|---|---|---|---|
| compare DALL-E 2 to SD v1.4 on woman mentions vs BLS | DALL-E 2 · Professions captions | % woman: 19.96% vs 38.04% (BLS 47.03%) | Hurt |
| SD v2 instead of SD v1.4 | Stable Diffusion · Professions captions | % woman: 38.04% -> 33.45% | Hurt |
| SD v2 instead of SD v1.4 | Stable Diffusion · Professions BLIP VQA 12 clusters | entropy (diversity): 2.30 -> 1.86 | Hurt |
| DALL-E 2 instead of SD v1.4 | TTI system · Professions BLIP VQA 12 clusters | entropy (diversity): 2.30 -> 1.36 | Hurt |
| DALL-E 2 instead of SD v1.4 | TTI system · Identities entropy BLIP VQA 12 clusters | entropy (diversity): 3.40 vs 3.35 | No clear effect |
| DALL-E 2 instead of SD v1.4 | TTI system · Woman-cluster share, Top 20% quintile | % images in woman clusters: 83.0 -> 56.3 | Hurt |
| DALL-E 2 instead of SD v1.4 | TTI system · Black-cluster share, Top 20% quintile | % images in Black clusters: 16.8 -> 4.5 | Hurt |
| use BLIP VQA embedding rather than CLIP embedding for identity clustering | audit method · Identities clustering | average entropy | Helped |
Closest papers by meaning
- Finetuning Text-to-Image Diffusion Models for Fairness (ICLR 2024 · similarity 0.90)
- OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes (ICLR 2025 · similarity 0.90)
- SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples (CVPR 2024 · similarity 0.90)
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.88)
- Leveraging Diffusion Perturbations for Measuring Fairness in Computer Vision (AAAI 2024 · similarity 0.88)
- Implicit Bias Injection Attacks against Text-to-Image Diffusion Models (CVPR 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.