Model Mixer
This is a study of famous women as trained into SDXL 1.0
I created a list of famous women's names and then I asked Stable Diffusion to generate an image of each woman. I then asked it to combine every combination of two women in the list into a set of images. You can scroll through these combinations. It's interesting that some women who I thought were famous, ended up being rendered as a pretty generic face.
Ethnicity Mixer
This is a study of ethnicities as trained into SDXL 1.0
I took the list of 722 ethnicities from the Global Ethnicities project on GitHub and then I asked Stable Diffusion to generate four images of women from each ethnicity. I then asked it to combine every combination of two ethnicities in the list into a set of four images. This site allows you to scroll through these combinations (or jump, with the random button).
That meant generating some 522,006 combinations, which will take a while. Each combination requires 4 images, and each image takes about 4.5 seconds, which gives me an estimated time to generate the whole set of about 109 days, if I generate 24/7. A best-case scenario would be generating about 6 hours a day, 7 days a week (basically while I'm sleeping), which means 435 days or well over a year (and is probably still unreasonable as only my main computer has the resources to run the images out at 4.5 seconds per image, or, really, at all).
So, instead I initially pared the list down to only 21 ethnicities, which required only 462 images, and now I've been able to extend that list to 101 ethnicities, manually curated, which required generating 41,208 images. The options are now much broader and the solution is more interesting (at least, in my opinion). As time allows I will continue to work on the full list of 722 ethnicities, however the time it will take to generate the whole 2,088,024 images is significant. Not to mention that I'll have to come up with a more robust engineering solution to storing over 2 million images on my web server in a way that won't be horrendously slow for Apache to serve the images (currently they are just in an images folder).
Therefore, if there is an ethnicity that you would like me to add sooner, rather than later, please let me know. Otherwise, I'll probably just add them in alphabetical order (see the full list of ethnicities at the github address above, or if you are willing to wait for 2,888 images to download, you can look at the complete list with representative images).
If there is anyone who would like to donate some computer time to generating the full list, or any other help, you would be most welcome. I could probably allow myself to accept bribes to adjust the priority of ethnicities as well.
Next Project here
Do you have any ideas for a next project? Suggestions