The last day or so I’ve been using Midjourney to create AI art. The reason is not that I’m artistic, I’m not really at all. Rather, I want to create illustrations for my Substack, and I figure it will be cheaper/easier than commissioning an artist. In fact, I already used something I generated with Midjourney for a Substack post.
But over the last hour, I aimed to produce a Western hunter-gatherer woman, and a Neolithic farmer man, and a Yamnaya pastoralist man.
Click for the larger image. It was really difficult to create the woman because the training inputs are such that when you stipulate that a woman has very dark skin, it just assumes you want to depict an African woman. So then I started adding “thin lips” to the prompt, and it turned the women into Europeans. My friend Askhar pointed out that you probably need to tell the program to start with an Indian woman, so that the dark-skinned feature doesn’t have strongly African features. Nevertheless, it was difficult to get the blue eyes to stand out on her face, so she ended up looking like she had Waardenburg syndrome. But I didn’t want to spend more than 30 minutes on it.
Here’s the prompt for her:
large blue eyes, cheek tattoos, face tattoos, long hair, black skin, beautiful woman, meadow background, shoulders visible, simple brown clothes, tribal Tamil woman, deep-set eyes, wide face, strong jaw, wide jaw, European characteristics, Hunter, primitive, face tattoos, cinematic, photorealistic, portrait –test –creative
The men were trivial to generate, so I’ll leave that to you.
Definitely some striking eyes on that HG woman.
I thought the blue eyes came from Ancestral North Eurasians. I’m surprised the Yamnaya didn’t have them – didn’t the eastern hunter-gatherers intermix with the ANE folk?
Very interesting results! I’m not sure if my impressions would be correct, but I’d’ve imagined the EEF man a bit more gracile and with sligthly finer and thinner features (I mean, thinner lips, narrower and more “pointy” nose), maybe just a little lighter in complexion, too (weren’t they basically in the same skin color range of the Yamnaya?). I had my own try, attempting to have a picture that represents credibly how I imagine the WHG women were, but you’re indeed correct in pointing out that it’s very hard to get an image with unquestionably blue eyes! Strange. So I tried to focus more on the other facial features. This is what I got, I’m not sure if it reliably fits the physical phenotypic range of WHG as estimated from their skulls (see link below). What do you think?
https://imgur.com/HXXIBtD
“Waardenburg syndrome.”
The spice must flow.
I don’t know if WHGs had large eyes.
For detailed information there is this paper
https://www.researchgate.net/publication/262985439_Craniometric_analysis_of_European_Upper_Palaeolithic_and_Mesolithic_samples_supports_discontinuity_at_the_Last_Glacial_Maximum
@Brett
ANE did not have blue eyes, blue eyes did not come from ANE. Blue eyes are first attested in the WHG Villabruna cluster based on the mutations that cause them today. The WHG Villabruna cluster is also the one with the highest frequency of blue eyes in the Bolling-Allerod to early and mid Holocene period.
“Ancestral North Eurasians.”
Maybe blondness did. Blue eyes are WHG.
For the really, really “best guess” way of doing portraits, I still think the best future solution will be learning from genotypes of existing people – whether that be PCA plus selected phenotype influencing genotypes for height, pigmentation etc or pure genotypes. As really for Yamnaya, for ex, it probably should be spitting out something further along the South European->North European vector (like 2-3 out), but with a less exaggerated height difference and darker pigmentation than present day Southern Europeans (due to those differences being amplified or created by selection over the last 5000 years). Probably some constraints against doing silly things too (like exaggerating some skeletal proportion beyond feasible limits). But AI with verbal cues doesnt really know any of that, so combines cues from existing photographed / depicted populations, in ways that give results that are pretty good for want of anything better, but will fall short of reality.
The technology probably exists already to easily do this, but it would take gathering the data, and the people with the resources don’t have such an interest in a big project to get an ultra perfect “best guess” phenotype.
Offtopic but:
48:15 – https://m.youtube.com/watch?v=o6axvZax1Ic – mentions an unpublished 100 genomes from Catal Hoyuk, Ankara group. This site will enter the age of large cemetery sequencing and hopefully characterisation of relationships beyond immediate family.
(Most of presentation is not new to anyone following adna).