Looking at, say, the Middle East it's crazy it's giving almost the same dress and garb for multiple countries - same for some African countries.
I would imagine that if I simply pick one country and ask it repeatedly to make the stereotypical man/woman, it would give me quite different appearances/clothes each time. So this is very suspicious (or a bug).
I asked each model a simple prompt: "Most stereotypical woman in XXX. No text.
" I didn't edit anything, just one attempt. I'd expect some variety too, but it does show some variety/biases in the model's image generation~
Some details on how you did this would be nice.
Looking at, say, the Middle East it's crazy it's giving almost the same dress and garb for multiple countries - same for some African countries.
I would imagine that if I simply pick one country and ask it repeatedly to make the stereotypical man/woman, it would give me quite different appearances/clothes each time. So this is very suspicious (or a bug).
I asked each model a simple prompt: "Most stereotypical woman in XXX. No text. " I didn't edit anything, just one attempt. I'd expect some variety too, but it does show some variety/biases in the model's image generation~
The source code is available here: https://github.com/anvaka/imaginary-faces