mlsanityData to 14 Oct 2026

Bias Amplification in Stable Diffusion's Representation of Stigma Through Skin Tones and Their Homogeneity

arXiv 2508.17465 ↗First posted 24 Aug 2025cs.CY cs.AIcited by 5

On the map: Diffusion models → Image generation, text-to-image and editing

Abstract (arXiv)

Text-to-image generators (T2Is) are liable to produce images that perpetuate social stereotypes, especially in regards to race or skin tone. We use a comprehensive set of 93 stigmatized identities to determine that three versions of Stable Diffusion (v1.5, v2.1, and XL) systematically associate stigmatized identities with certain skin tones in generated images. We find that SD XL produces skin tones that are 13.53% darker and 23.76% less red (both of which indicate higher likelihood of societal discrimination) than previous models and perpetuate societal stereotypes associating people of color with stigmatized identities. SD XL also shows approximately 30% less variability in skin tones when compared to previous models and 18.89-56.06% compared to human face datasets. Measuring variability through metrics which directly correspond to human perception suggest a similar pattern, where SD XL shows the least amount of variability in skin tones of people with stigmatized identities and depicts most (60.29%) stigmatized identities as being less diverse than non-stigmatized identities. Finally, SD shows more homogenization of skin tones of racial and ethnic identities compared to other stigmatized or non-stigmatized identities, reinforcing incorrect equivalence of biologically-determined skin tone and socially-constructed racial and ethnic identity. Because SD XL is the largest and most complex model and users prefer its generations compared to other models examined in this study, these findings have implications for the dynamics of bias amplification in T2Is, increasing representational harms and challenges generating diverse images depicting people with stigmatized identities.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Placed in Image generation, text-to-image and editing, one of 14 directions in Diffusion models. Builds on 22 of its 52 references that are in the map. First posted on arXiv 24 Aug 2025.

Builds on

22 of 52 references.

Closest ancestors · 2

Same direction, newest first

  1. 2023'Person' == Light-skinned, Western Man, and Sexualization of Women of Color: Stereotypes in Stable Diffusion30 Oct
  2. 2022DALL-Eval: Probing the Reasoning Skills and Social Biases of Text-to-Image Generation Models8 Feb

Other references · 20

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesiscited by 4,356
  3. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,152
  4. 2017Progressive Growing of GANs for Improved Quality, Stability, and Variationcited by 2,988
  5. 2020Language (Technology) is Power: A Critical Survey of "Bias" in NLPcited by 956
  6. 2019MaskGAN: Towards Diverse and Interactive Facial Image Manipulationcited by 628
  7. 2019Towards a Critical Race Methodology in Algorithmic Fairnesscited by 125
  8. 2024A Note on Shumailov et al. (2024): `AI Models Collapse When Trained on Recursively Generated Data'cited by 115
  9. 2021Towards Measuring Fairness in AI: the Casual Conversations Datasetcited by 80
  10. 2020Lifespan Age Transformation Synthesiscited by 60
  11. 2023Bias Against 93 Stigmatized Groups in Masked Language Models and Downstream Sentiment Classification Taskscited by 51
  12. 2022Contrastive Language-Vision AI Models Pretrained on Web-Scraped Multimodal Data Exhibit Sexual Objectification Biascited by 43
8 more
  1. 2022Towards Racially Unbiased Skin Tone Estimation via Scene Disambiguationcited by 32
  2. 2024Large Language Models Portray Socially Subordinate Groups as More Homogeneous, Consistent with a Bias Observed in Humanscited by 31
  3. 2022Evidence for Hypodescent in Visual Semantic AIcited by 30
  4. 2023A Friendly Face: Do Text-to-Image Systems Rely on Stereotypes when the Input is Under-Specified?cited by 25
  5. 2023Beyond Skin Tone: A Multidimensional Measure of Apparent Skin Colorcited by 17
  6. 2024Vision-Language Models Generate More Homogeneous Stories for Phenotypically Black Individualscited by 6
  7. 2021Black or White but never neutral: How readers perceive identity from yellow or skin-toned emojicited by 3
  8. 2025Documenting Patterns of Exoticism of Marginalized Populations within Text-to-Image Generatorscited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026VEAT Quantifies Implicit Associations in Text-to-Video Generator Sora and Reveals Challenges in Bias Mitigationcited by 2
  2. 2026Toward Trustworthy Portrait Editing: Evaluation of Demographic Misrepresentation in I2I Modelscited by 1
  3. 2026Position: the Stochastic Parrot in the Coal Mine. Model Collapse is a Threat to Low-Resource Communitiescited by 1
  4. 2026Gender Artifacts from Art History to Text-to-Image Generationcited by 1
  5. 2026GLEaN: A Text-to-image Bias Detection Approach for Public Comprehensioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024AI-generated faces influence gender stereotypes and racial homogenization
  2. 2025Visual Cues of Gender and Race are Associated with Stereotyping in Vision-Language Models
  3. 2023RS-Corrector: Correcting the Racial Stereotypes in Latent Diffusion Models
  4. 2025Hidden Bias in the Machine: Stereotypes in Text-to-Image Models
  5. 2023Leveraging Diffusion Perturbations for Measuring Fairness in Computer Vision
  6. 2024More Distinctively Black and Feminine Faces Lead to Increased Stereotyping in Vision-Language Models
  7. 2022Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scale
  8. 2026Measuring Social Bias in Vision-Language Models with Face-Only Counterfactuals from Real Photos