shaghayegh/stylistic-bias-dataset

Link to the paper

https://arxiv.org/abs/2606.20527

Dataset Description

StylisticBias is synthetic controlled benchmark for evaluating fine-grained, attribute-level social bias in multimodal large language models (MLLMs). It addresses a key limitation of prior work: most bias studies compare different individuals or demographic groups, making it impossible to disentangle appearance-driven effects from identity-level variation.

StylisticBias keeps identity fixed and varies one visual attribute at a time, enabling direct measurement of how specific visual cues — such as hair style, eyewear, tattoos, or fashion style — shift an MLLM's social judgments about a person.

• Paper: StylisticBias: A Few Human Visual Cues Drive Most Social Biases in MLLMs
• https://arxiv.org/abs/2606.20527
• Authors: Shaghayegh Kolli*, Timo Cavelius* (Equal Contribution), Nafiseh Nikeghbal, Samantha Dalal, Jana Diesner
• Code: github.com/timo-cavelius/StylisticBias
• License: CC-BY 4.0
• Languages: English (evaluation prompts and scenario labels)
• Modalities: Image + Text

---

Dataset Summary

Property	Value
Total images	~25,000
Base faces	500 synthetic photorealistic identities
Variations per face	~50 single-attribute edits
Demographic configurations	90 (3 age × 2 gender × 5 ethnicity × 3 body type)
Attribute categories	12 (skin, hair color, hair length, hair style, facial hair, makeup, lip makeup, eyewear, piercings, tattoos, headwear/accessories, fashion style)
Individual attribute values	~55 (full grid); 34 used in evaluation after reduction
Splits	train / validation / test: 20,030 / 2,492 / 2,515 rows (400 / 50 / 50 base faces); each face and all its variations are in exactly one split
Image resolution	1024×1024 px
Format	Parquet (with embedded images)

---

Dataset Structure

Columns

Column	Type	Description
image	image	The generated variation image (for base rows: the base face itself)
face_id	int32	Number of the person, 1–500. The base face and all its variations share it. Train: 1–400, validation: 401–450, test: 451–500
variation_id	int32	Which edit the image shows: 0 = the unedited base face, 1–58 = one attribute edit. The same number means the same edit on every person (see Variation Codes)
id	string	Unique identifier for the record
face_group	string	ID of the base face (identity) this row belongs to
asset_kind	string	Type of image: base (the unedited base face), face (head-and-shoulders variation), body (full-body, for fashion variations)
image_file	string	Filename of the variation image
image_relative_path	string	Relative path to the variation image within its face group folder
metadata_relative_path	string	Relative path to the JSON metadata file for this variation
age	string	Age group of the base face: young adult, middle-aged adult, elderly
gender	string	Perceived gender of the base face: male, female
ethnicity	string	Ethnicity label of the base face: Asian, African, European, Middle Eastern, Latino
body_type	string	Body type of the base face: thin, normal, obese
variation_attribute	string	The single attribute varied in this image (e.g., hair_style); empty for base rows
variation_value	string	The value of that attribute (e.g., Messy); empty for base rows
prompt	string	The exact generation prompt used to produce this image
timestamp	string	ISO 8601 timestamp of image generation
base_image_name	string	Filename of the base (reference) face image; equals the image_file of the face's base row

Asset Kinds

• base: The unedited base face of an identity (one row per face, 500 in total). It has variation_id 0 and is always the first row for that person.
• face: Head-and-shoulders portrait. Used for all attribute categories except fashion (skin irregularities, hair, makeup, eyewear, piercings, tattoos, accessories).
• body: Full-body portrait. Used exclusively for fashion style variations, which require a full-body view to be meaningful.

How Base Faces and Variations Are Linked

Every row carries two numbers:

• face_id identifies the person. The base face and all its variations share the same face_id.
• variation_id identifies the edit. 0 is the unedited base face; the same variation_id means the same edit on every person (e.g. 6 = black hair).

Rows are sorted by face_id, then variation_id: each person's base face comes first, followed by their variations in the same order for everyone. For example, face_id 17, variation_id 0 is the original of face_id 17, variation_id 6.

from datasets import load_dataset

ds = load_dataset("shaghayegh/stylistic-bias-dataset", split="test")

# Row of each person's base face (reads only the number columns, no image decoding)
base_row = {f: i for i, (f, v) in enumerate(zip(ds["face_id"], ds["variation_id"])) if v == 0}

row = ds[1]                          # a variation
base = ds[base_row[row["face_id"]]]  # the same person's unedited face

# The same edit across all people, e.g. black hair
black_hair = ds.filter(lambda v: v == 6, input_columns="variation_id")

For every person, face_id + variation_id identifies exactly one image, except for 16 people who have two images with variation_id 45 (Hat), generated twice from the same prompt. Values with two spellings in variation_value share one variation_id. face_group and base_image_name hold the original text identifiers.

Variation Codes

All 59 codes, with the number of faces that have each edit

variation_id	variation_attribute	variation_value	Faces
0	—	— (unedited base face)	500
1	skin_irregularities	Freckles	498
2	skin_irregularities	Acne	499
3	skin_irregularities	Scars	499
4	skin_irregularities	Moles	494
5	skin_irregularities	Birthmarks	3
6	hair_color	Black	499
7	hair_color	Brown	498
8	hair_color	Blonde	499
9	hair_color	Red	499
10	hair_color	Gray	499
11	hair_color	Unnatural color (also spelled Unnatural colors)	498
12	hair_length	Bald	498
13	hair_length	Short	498
14	hair_length	Medium	498
15	hair_length	Long	498
16	hair_style	Messy	498
17	hair_style	Slicked back	498
18	hair_style	Ponytail	498
19	hair_style	Braid	401
20	hair_style	Bun	401
21	hair_style	Afro	498
22	hair_style	Buzz cut	498
23	hair_style	Mohawk	498
24	facial_hair_male	Clean-shaven	267
25	facial_hair_male	Stubble	267
26	facial_hair_male	Mustache	267
27	facial_hair_male	Full beard	267
28	eyewear	Thick-rimmed	498
29	eyewear	Thin metal	498
30	eyewear	Sunglasses	498
31	makeup_female	Light	219
32	makeup_female	Heavy	219
33	lip_makeup_female	Neutral lipstick	218
34	lip_makeup_female	Red lipstick	218
35	lip_makeup_female	Bold color (also spelled Bold colors)	218
36	piercings	Single nose	496
37	piercings	Single lip	496
38	piercings	Single eyebrow	496
39	piercings	Multiple	498
40	piercings	Earrings (also spelled Earring)	498
41	piercings	Single	2
42	tattoos	Facial tattoo	497
43	accessories	Cap	497
44	accessories	Beanie	496
45	accessories	Hat	496
46	accessories	Headscarf	493
47	accessories	Veil / headscarf	2
48	fashion_style	Professional / Business formal	496
49	fashion_style	Formal / Evening wear	495
50	fashion_style	Casual	495
51	fashion_style	Smart casual	495
52	fashion_style	Sporty / Athletic wear	495
53	fashion_style	Streetwear	495
54	fashion_style	Functional / outdoor wear	494
55	fashion_style	Luxury / High fashion	492
56	fashion_style	Vintage / Retro	487
57	fashion_style	Worn / Distressed clothing	487
58	fashion_style	Daring / Provocative	202

Facial hair exists only for male faces, makeup and lip makeup only for female faces. Codes 5, 41 and 47 come from a few early generations with a different prompt wording.

Earlier layout (v1.0). Until October 2026, every row also stored its base face in a base_image column (61.6 GB in total). That version stays available with load_dataset("shaghayegh/stylistic-bias-dataset", revision="v1.0").

Attribute Space

The full variation space is as follows. Values excluded from the evaluation subset (but present in the dataset) are marked with *.

Attribute	Values
Skin irregularities	Freckles, Acne, Scars*, Moles
Hair color	Black, Brown, Blonde, Red*, Gray, Unnatural color
Hair length	Bald, Short, Medium*, Long
Hair style	Messy, Slicked back, Ponytail*, Braid*, Bun*, Afro*, Buzz cut*, Mohawk
Facial hair (M only)	Clean-shaven, Stubble*, Mustache*, Full beard
Eyewear	Thick-rimmed, Thin metal*, Sunglasses
Makeup (F only)	Light, Heavy
Lip makeup (F only)	Neutral*, Red lipstick, Bold*
Piercings	Single nose, Single lip*, Single eyebrow*, Multiple, Earrings*
Tattoos	Facial tattoo
Accessories	Cap, Beanie, Hat*, Headscarf*
Fashion style	Professional/Business formal, Formal/Evening wear, Casual, Smart casual, Sporty/Athletic wear, Streetwear, Functional/outdoor wear, Luxury/High fashion*, Vintage/Retro, Worn/Distressed clothing, Daring/Provocative*

*Excluded from the evaluation subset for plausibility or redundancy reasons (see paper Appendix C.1).

---

Dataset Creation

Source Data

All images are fully synthetic — no real photographs of individuals were used. Base faces were generated using Imagen 4 (Google DeepMind, 2025). Attribute variations were produced using Nano Banana (Gemini 2.5 Flash Image) (Comanici et al., 2025).

Base Face Generation

500 base faces were generated from a structured demographic prompt family. The Cartesian product of four demographic attributes — age (young adult, middle-aged, elderly) × gender (male, female) × ethnicity (Asian, African, European, Middle Eastern, Latino) × body type (thin, normal, obese) — yields 90 unique configurations, from which 500 identities were sampled for broad demographic coverage.

Base prompts enforce: front-facing pose, neutral expression, head-and-shoulders framing, plain white background, soft even lighting, no accessories or makeup, natural skin texture.

Observed base-face distribution:

• Gender: 274 male, 226 female
• Body type: 186 normal, 160 obese, 154 thin
• Ethnicity: 110 Asian, 109 African, 101 European, 95 Middle Eastern, 85 Latino
• Age: 260 young adult, 124 middle-aged, 116 elderly

Variation Generation

For each base face, single-attribute variations were generated by modifying exactly one feature at a time while explicitly preserving the base identity. Fashion-style variations use a full-body prompt template; all other variations use a head-and-shoulders portrait template.

Human Validation

90% of all generated images (base faces and variations) were manually reviewed, checking for: demographic plausibility, identity consistency across variations, and correct realization of the intended attribute change without unintended artifacts. 98% of reviewed images passed validation; failures were regenerated and re-reviewed before inclusion.

---

Intended Use

StylisticBias is intended as a benchmark for bias evaluation in multimodal large language models. It is designed for researchers studying:

• Attribute-level visual bias in vision-language models
• How specific appearance cues shift model social judgments
• Intersectional effects of demographic context and visual attributes
• Fairness auditing of deployed multimodal AI systems

Out-of-Scope Use

This dataset is not intended for:

• Re-identification of real individuals (all faces are synthetic)
• Any application that would operationalize the social judgment scenarios evaluated here (e.g., automated hiring or judicial support systems)

---

Evaluation Setup

In the companion paper, six open-source MLLMs were evaluated on this benchmark:

Model	Provider	Params
LLaVA-v1.6-Mistral-7B	LLaVA Team	7B
Qwen3-VL-8B-Instruct	Alibaba	8B
Pixtral-12B	Mistral AI	12B
InternVL3-14B	OpenGVLab	14B
Gemma-3-12B-IT	Google DeepMind	12B
Gemma-4-E4B-IT	Google DeepMind	4B (effective)

Models were queried on 25 binary social judgment scenarios spanning four categories: Personality & Social, Interpersonal, Behavioral, and Socioeconomic & Appearance. Each (image, scenario) pair was evaluated under 4 prompt orderings × 3 random seeds = 12 prompts, for approximately 4.72 million judgment calls per model and ~28.3 million total.

Limitations

1. Synthetic images only. The benchmark uses controlled synthetic data rather than real photographs. This is a deliberate design choice to enable single-attribute controlled variation, but the benchmark may not capture the full distribution of real-world photographs. Conclusions characterize model behavior in a controlled visual setting.

2. Curated demographic and attribute scope. The study covers a focused set of demographic groups and visual attributes. Many other identities and appearance dimensions exist and are not represented. The broad demographic categories used (e.g., ethnicity labels) are social constructs and may not reflect the full complexity of real-world identity.

3. Input-level effects only. The benchmark measures how visual attributes shift model outputs; it does not explain the underlying mechanisms producing these effects.

---

Ethical Statement

This dataset studies how visual attributes drive social judgments in MLLMs deployed in consequential settings. All faces are fully synthetic and do not represent or resemble any real individual. Synthetic face generation reduces privacy risks but may reproduce stereotypical associations from the generative models' training data.

We acknowledge dual-use risks: the methodology used to construct this benchmark could in principle inform adversarial appearance manipulation in automated pipelines. We release StylisticBias specifically to support fairness auditing and bias attribution research.

The demographic categories and social judgment scenarios used in this benchmark are social constructs that can reflect stereotypical perceptions and normative expectations. They are operationalized here strictly for the purpose of measuring and quantifying bias in model behavior, not to endorse or reproduce those judgments.

---

Citation

@misc{stylisticbias2026,
  title={StylisticBias: A Few Human Visual Cues Drive Most Social Biases in MLLMs},
  author={Shaghayegh Kolli and Timo Cavelius and Nafiseh Nikeghbal and Samantha Dalal and Jana Diesner},
  year={2026},
  eprint={2606.20527},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2606.20527}
}

Dataset Card Authors

Shaghayegh Kolli*, Timo Cavelius* (Equal Contribution), Nafiseh Nikeghbal, Samantha Dalal, Jana Diesner