This dataset was created as part of a research collaboration project that seeks to reduce the AI devide for marginalized communities by improving the representation of people with disabilities in text-to-image model outputs. It brings together two types of datasets: (1) a Community Library of real photos with corresponding meta-data text annotations; and the related (2) Community Preference dataset, which is synthetic, AI-generated dataset that is intended to capture a community's preffered representation.

(1) Community Library

This dataset was developed through a research collaboration project aimed at reducing the AI divide for marginalized communities by improving the representation of people with disabilities in text-to-image model outputs. The dataset focuses on the short stature community, featuring 400 real-world images with corresponding metadata. The images highlight individuals and groups, reflecting key representation themes and subthemes defined by the community as “good representation.” They also capture diversity (such as categories of dwarfism), different age groups (such as children, adults), a range of activities (such as sports, work), and varied settings (such as school, home, farms). Collected in Kenya, the dataset was curated between April and July 2025.

(2) Community Preference Dataset

The Community Preference dataset is a synthetic dataset that is intended to capture a community’s preferred representation. It consists of 315 rows, each containing: 1) reference image ID (where applicable), 2) reference prompt (where applicable); 3) extra prompt (reference prompt augmented with meta-data); 4) AI-generated image; 5) a rating of alignment to preferred representation by the community lead between 1-5; and 6) name of model generating image.

Distribution by Generation Model and Category

Model	Diversity	Education & Hobbies	Relationships	Sports	Work	Generic Prompts	Total
GPT-1	15	15	15	15	15	40	115
Imagen-4	15	15	15	15	15	40	115
SD-3.5	15	15	15	15	15	10	85
Total	45	45	45	45	45	90	315

Supported Tasks

Text-to-Image: The dataset can be used to train, evaluate or modify text-to-image generation models.

Evaluation: The Community Preference dataset is best suited for training community-specific evaluator models with the goal of improving disability representation in generative AI media.

Languages

The annotations within each image are in English. The associated BCP-47 code is en.

Dataset Structure

Data Instances

For the Community Library dataset, this is what a typical JSON-formatted example looks like): { "theme_folder": { "name": "theme_name", "description": "theme_description", "sub_themes": { "sub_theme_folder": { "name": "sub_theme_name", "description": "sub_theme_description", "images": [ { "path": "path-to-file", "description": "image_description", "prompt": "image_prompt", "annotations": [ { "label": "label_for_box", "box": { "x": 0.25, "y": 0.25, "w": 0.5, "h": 0.5 } } ] }, ] } } } } For the Community Preference dataset, this is what a typical JSON-formatted example looks like): { “ref_image_id”:””, "image_id": "", "prompt": "", “extra_prompt”: “” "model": "", "score”: “”} There is an additional community meta-data file called info.json with the items “name”, “shortname”, “disability_type”, and “description”.

Data Fields

(1) Community Library

The meta data includes: (1) a hierarchy of themes (with name and description) and subthemes (title, name, description) that reflect desired representation aspirations of the community across the entirety of the 400-image dataset. For each image (path) in the dataset, the meta-data further includes: (2) a rationale for “why” an image has been selected as a good instance of representation theme (description); (3) a prompt describing the image (prompt); and (4) 1-5 image text (label) and bounding-box annotations via x, y, w, h dimensions.

(2)Community Preference Dataset

The meta data includes: the “ref_image_id”, which references the real-world image counter-part from the Community Library; the “imageid”, which is the unique identifier for the generated image; the “prompt”, a text description of the corresponding reference image in the community library; the “extra_prompt”, additional text about an image location and its highlights that gets concatenated to the “prompt” to generate that image; and the “model”, which indicates which AI model was used to produce the image.

Dataset Creation

Curation Rationale

(1) Community Library

The dataset was curated to support the adaptation and evaluation of text-to-image generative models. It is organized around five core themes, chosen through participatory input from the community to capture meaningful activities, characteristics, and objects for AI generated imagery. Each theme is sub-divided into hierarchical sub-themes, offering a structured taxonomy. Within each sub-theme, curated images and detailed annotations illustrate the visual and semantic traits of the category. This structure enables targeted evaluation of model performance across conceptual domains and facilitates research on theme-specific generation fidelity, compositional generalization, and prompt grounding.

(2) Community Preference Dataset

The dataset was created to support the development and validation of a scalable metric of preferred representation, grounded in the Community Library. In general, the aim of the dataset is to provide a reliable, community-aligned signal for assessing how well AI-generated images reflect the community’s preferred ways of being represented, and to enable consistent comparison across models, prompts, and settings.

Source Data

Initial Data Collection

(1) Community Library

The dataset was collected between June and July 2025 using a participatory, community-driven approach. Participants were persons of short stature, selected to ensure diversity in age, gender, categories of dwarfism, and social roles. Images were captured in various real-world settings including homes, schools, workplaces, farms, sports fields and community events across multiple Kenyan regions. Geographical and contextual variety ensured broad representation. The review process prioritized image quality, clarity, relevance to pre-defined themes, and accurate depictions of activities, objects, and characteristics valued by the community. Final inclusion emphasized diversity, cultural authenticity, and meaningful representation within each thematic category.

(2) Community Preference Dataset

The Community Preference dataset is created by generating images from 100 text-prompts. The majority of these (n = 80) are derive from captions of real photographs in the Community Library. Editable text-prompts are automatically generated based on a community member response to the question, ‘why is this image a good representation of your community?’ as well as the real image, which are given as inputs to GPT-4o. These text-prompts are systematically sampled across all community-identified themes and subthemes.

The other set of prompts (n=20) are intended to reflect prompts that might be entered by a general user looking to generate images of a disability community. These prompts were derived by domain experts at Microsoft inspired by a set of shared scenarios / use cases for image generation models which are important to disability communities. They are often much shorter than the prompts derived from the Community Library. Each of the 100 prompts is used to generate an image from each of three models: GPT Image-1; Imagen-4Ultra; and Stable Diffusion 3.5 Large turbo. These models were chosen to provide a spectrum of ‘good to bad’ representation needed for solid metric development. The data was generated in June/July 2025.

Who are the source data producers?

(1) Community Library

The dataset was produced entirely by humans. The images were originally created by local photographers and community volunteers who collaborated with persons of short stature to document authentic representations. Demographic information about the image creators is not recorded and remains unknown. Data was generated through two primary approaches: direct image capture by photographers and voluntary data donations from community members who contributed personal or family photographs. All contributors participated willingly, with informed consent obtained before inclusion. To recognize their time and effort, photographers and community members who provided images or assisted in curation received fair monetary compensation.

(2) Community Preference Dataset

The generation of the images of this dataset was led by the Microsoft research team.

Annotations

Annotation process

(1) Community Library

(a)Dataset-level annotations - To structure their image library, annotators grouped and categorized images into 5 themes (e.g., Work, Relationships, Sports, Diversity and Education/Hobbies) with 3 sub-themes (e.g., formal profession, informal profession, family relationship, romantic relationship, competitive single person sport, competitive team sport e.t.c).

(b)Individual image-level annotations - Instructions for the creation of the annotations for each image involved the annotators’ response to the following questions/ tasks: Why did you select this image as a good representation for your community? Edits to an auto-generated prompt to ensure its accuracy in regards to the image and in describing what is important for the community using their preferred language. A set of 1-5 bounding box annotations that either relate to: (i) Objects that are special or specific to the community (e.g., Step Stool, Adapted Car); or (ii) People that are important for your community (e.g., Young business woman of short stature).

(2) Community Preference Dataset

A carefully designed rating tool was used to provide an ordinal rating of 5 steps from 1 (very bad) to 5 (very good) that holistically communicates how well each generated image aligns to the community’s preferred representation given the text-prompt from their Community Library.

Who are the annotators?

All annotations were completed by a single individual who served as the project lead for the dataset generation process. No demographic or identity information about the annotator is provided and therefore remains unknown.

For the Community Library, annotations were created under controlled conditions to ensure consistency, aligning with the predefined themes and subthemes that guided dataset organization. The annotator reviewed each image for quality, thematic relevance, and representation before assigning metadata. The same individual provided all annotations for the Community Preference Dataset. As part of the project’s ethical framework, fair monetary compensation was provided for the time and expertise dedicated to the annotation process.

Personal and Sensitive Information

(1) Community Library

The dataset is not explicitly linked to individuals, and no personal identifiers are included and therefore subjects cannot be directly identified. However, images may reveal sensitive information such as racial and ethnic origins, cultural backgrounds, or visible health related characteristics associated with specific categories of dwarfism. The dataset also contains images of children participating in various activities. While all images were collected with informed consent, users should be aware of the potential for indirect sensitivity in certain contexts and handle the dataset with appropriate ethical considerations and privacy safeguards.

(2) Community Preference Dataset

All AI generated, synthetic data contained in this dataset shows images of people, including those with disabilities and including children. All data has been checked by the project lead and the research team for likeness to the real images in the Community Library, since it cannot be fully excluded that a face is copied.
Furthermore, some AI image generation models produce offensive images. We deemed the following offensive: 1) total erasure of disability; 2) spurious correlations (e.g. putting pointy ears or a beard on someone with dwarfism); and 3) any violent or sexualized content. A manual content filtering stage was used to ensure that community project lead was not exposed to offensive images. A particular prompt was generated up to 10 times to attempt to get a non-offensive image. If one could not be produced, an image from the 10 examples was randomly chosen (assuming no sexual or violent content). As such, there is still imagery in this dataset that some may find offensive.

Considerations for Using the Data

Social Impact of Dataset

(1) Community Library

The dataset has the potential to create significant positive social impact by improving the representation of persons of short stature in AI-generated imagery, fostering inclusion, and reducing the AI divide for marginalized communities. The dataset can support the development of technologies that generate fairer, more respectful, and diverse depictions, contributing to greater social awareness and positive cultural narratives. Such advancements may enhance accessibility, representation in media, and equitable participation in emerging AI-driven industries. However, its use also carries risks. Misuse of dataset could enable the creation of stigmatizing, defamatory, or pornographic content that harms individuals or the community it represents. It may also be exploited for surveillance, discriminatory profiling, or the reinforcement of harmful stereotypes. Despite the dataset not containing explicit personal identifiers, privacy risks remain, including potential re-identification of individuals. To mitigate harm, users must adhere to ethical guidelines, prioritize responsible use, and ensure outputs do not perpetuate discrimination or bias.

(2) Community Preference Dataset

The Community Preference dataset provides a foundation for advancing both methodological and applied research on community-aligned AI evaluation. The data enables the development and validation of new evaluator models, supporting innovation in how these are constructed, trained, and validated against community ratings. As a resource, this dataset and any evaluator models built from it, can lower the barrier to entry for studying participatory, community-grounded evaluation, fostering reproducibility and enabling more systematic progress on representation-aware assessment in generative AI models.

Discussion of Biases

(1) Community Library

The dataset may reflect inherent biases due to its scope and collection context. It primarily features persons of short stature from Kenya, limiting geographic, cultural, and ethnic diversity. Certain themes, sub-themes, settings, or activities – such as experiences of children, the elderly, or individuals in less-documented environments – maybe underrepresented. Additionally, the dataset focuses exclusively on one disability community rather than encompassing a broader spectrum of disabilities, which may narrow its generalizability. To reduce these impacts, participatory input guided theme selection, and efforts were made to include diverse age groups, activities and settings. Future expansions could address these gaps by incorporating additional regions, broader disability categories, and more varied life contexts to enhance representational balance. Limitations:The dataset is relatively small in scale and may not capture the full diversity of lived experiences. Its focus on Kenyan contexts limits global applicability. It is intended primarily for research and model evaluation, not as a definitive or exhaustive representation of all persons of short stature.

(2) Community Preference Dataset

The Community Preference dataset was developed for research and experimental purposes. It reflects what state-of-the-art AI image generation models could do at the time of creation. Given how rapidly this field is moving, this is likely to be very different in a few years.

This dataset has not been systematically evaluated for sociocultural/ economic/ demographic/ linguistic bias. Developers should consider the potential for bias as they select use cases, and evaluate and mitigate for accuracy, safety, and fairness concerns specific to each intended downstream use. It should not be used in highly regulated domains where inaccurate or incomplete outputs could suggest actions that lead to injury or negatively impact an individual's legal, financial, or life opportunities.

Additional Information

Dataset Curators

The dataset was curated by Ruth Mueni and Short Stature Society of Kenya in collaboration with Microsoft Research. This partnership combined community led insights with technical expertise to ensure authentic representation, ethical data handling, and alignment with research goals focused on improving inclusivity in text-to-image generative models.

Licensing Information

This dataset is licensed under the Creative Commons Attribution-ShareAlike 4.0 license.

Contributions

We extend heartfelt gratitude to the community members whose participation made this dataset possible. Special appreciation goes to Mike Odera of the Short Stature Society of Kenya for his invaluable support in the dataset curation/open-sourcing efforts, helping ensure the dataset reflects the lived experiences and voices of the community as well as their authentic representation and meaningful contributions to inclusive AI research.

Downloads last month: 28

Total file size:

190 MB