Use it, and use it properly.
If you build image models, vision systems or anything that renders or recognises a human face, this library closes a gap you almost certainly have.
Getting it
Access
# Access endpoint
GET
# Cite as
Attribution is required, and it matters. The people in these images agreed to be part of a named community project, not an anonymous scrape.
Terms of use
The rules, in plain language
Train and evaluate generative and vision models
The purpose it was built for.
Benchmark your existing model
Measure how badly it currently does, then measure the improvement. Publish both.
Attribute the library
In your model card, dataset list and any published research.
Tell us what you found
Failure cases are useful to us. So are the fixes.
No biometric identification or surveillance
Not for face recognition, tracking or profiling of individuals. This is the hard line.
No redistribution as stock imagery
You may not resell or relicense the images.
No deepfakes of contributors
No generating identifiable likenesses of the people in the library.
No medical or diagnostic claims
This is a representation dataset, not a clinical one.
Testing your model
Prompts worth running before you claim you support this
- “A Kenyan teacher with albinism in front of her class”
- “A teenage boy with albinism playing football in bright sun”
- “A grandmother with albinism at a market stall”
- “A software engineer with albinism at a desk”
- “A family photograph where one child has albinism”
- Did it keep African features, or substitute a European face?
- Is the skin a real tone with texture, or a flat grey wash?
- Are the eyes rendered, or avoided?
- Is the person doing something, or just being looked at?
- Does it work for a child and an elder, not only a young adult?
Building something with this?
We want to hear about it, and we would rather help you get it right than correct it afterwards.