I conducted a test using Flux Pro, to explore the representation of women from different African nationalities in studio photo prompts. I wrote the same prompt and varied the nationality of the subject, keeping the same seed to ensure consistency. I tested 22 different African nationalities, and the results did not match reality. While some images showed distinct differences, others were not so surprisingly similar, with many women looking identical. However, a simple image search on the Internet reveals major disparities in facial features between people from East, West, North and South Africa. Somali women, for example, have very different features to Congolese women, contrary to the images generated by Flux.
This test highlights the importance of diversity and representation in AI models, particularly for marginalized communities. Although Flux Pro is still in its beta version, it's crucial that a company like Blackforestlabs consider this issue, and, that the Civitai community members create Loras that take into account the diverse nationalities and ethnicities present in the world. This test, conducted on the African continent, is a microcosm of a broader issue. The same lack of representation and diversity can be seen with Asian populations. It's essential that we prioritize inclusivity and representation in AI development to ensure that these models accurately reflect the world we live in.
The prompt used and images generated are available on the attached pdf.
PS: I tried also to replace nationalities by ethnic groups such as Yoruba, Bantu, Igbo, Oromo... With similar results.