[Content warning: pixels representing paint representing breasts]
My favorite thing about AI art is that the results are throw-aways. You are not asking a real artist to pour weeks of work and years of skill into something that is, basically, a joke.
In another window I’m brain-farting over a version of Stable Diffusion that is crunching away at a nude version of the Mona Lisa. Personally, Mona Lisa is hardly one of my favorite works – it’s small, dark, dour, uninteresting – but “we can fix that!” is the motto of the AI artist. I’ll update this posting with a better version if I get one I really like:
Stable Diffusion AI and mjr: “mona lisa overdrive”
I normally would have used Midjourney but dicks and breasts and stuff like that are not allowed, because offense or something. So I turned to one of my other favorites and asked for Midjourney to goth it up.
Midjourney AI and mjr “whistler’s mother as an 80s goth chick with big hair and lots of makeup.”
I legitimately love that image. The character of the image is completely changed, yet it still resembles its original. Fascinating. Look what it did with the background painting!
Midjourney AI and mjr: “whistler’s mother as an 80s goth”
Meanwhile, another favorite of mine, “Madame X” by John Singer Sargent. What a wonderful painting for a goth remix. So, I gotta confess something – one of my benchmarks of beauty is a good nose. Madame X’s nose is spectacular. Midjourney kind of snubbed it down. But I can ask for profile views, etc. This came out pretty good:
Midjourney AI and mjr, “Madame X as an 80s goth chick”
Actually, a lot of them came out spectacular. But I can’t bury you with them.
Midjourney AI and mjr “Madame X as an 80s goth chick”
I know that variations on Grant Woods’ otherwise boring “American Gothic” have been done to death, but I just did one for the hell of it.
Midjourney AI and mjr, “American re-goth”
It didn’t understand “just the woman” and redecorated the guy, too. Kind of a bummer, really, but AIs work that way – once you have activated “mohawk hairdo” in the prompt it tries to mohawk all the hair.
To produce these, I provided the original as a prompt reference, which tends to limit the AI’s ability to go far afield. My text prompt asked for torn mesh stockings and a pinned tshirt for Madame X but it couldn’t leave the elegance of the image.
This is Stable Diffusion’s version of the Grant Wood:
I’d love to see what you/Midjourney/Stable Diffusion could do with my favourite painting in the world, Amaury-Duval’s portrait of Marie-Anne de Tourbay, https://en.wikipedia.org/wiki/Marie-Anne_Detourbay
I saw her as I was about to leave the Musee D’Orsay. I’ve stood for hours in front of all kinds of art, and felt all sorts of things, but nothing and nobody has ever seen straight into the centre of me and DEMANDED I come over and look at her the way that painting did. I don’t really know how long I was there. I had to be almost dragged out of there, I know that much. If I could own one piece of art, one object of any kind come to that, it would be that painting. It’s unsettling how good it is. No reproduction does it justice (although I could say the same of Nighthawks or any number of other paintings I’ve seen “in the flesh” – quite a few of the best are just not as good in any kind of reproduction. The ultimate example of this is “The Fairy Feller’s Master Stroke”, which is a very strange experience too. Can’t even imagine what AI “artists” would make of trying to come up with something “like” that, because there’s really nothing like it, nor could there be.
Rambling, sorry.
If you remember an old Maxell tape advert (which may never have aired in the US?) you might like this:
http://www2.b3ta.com/host/creative/72057/1403872878/tape2mp3MaxellAd.gif
American Gothic (first version) – that’s a useless looking pitchfork.
I just now noticed that the pitchfork form is duplicated in the man’s clothing, the seams of the overalls and the lines of the shirt. This is true to the original.
I see Mona is a victim of polydactyly.
@sonofrojblake:
Reginald Selkirk@#3:
I see Mona is a victim of polydactyly.
I fixed it. De Caprio edited her fingers in the original, and removed the gills from her neck, too.
@mjr,4: crikey. Thank you.
I spend way too much time just putting cute animals into turtleneck sweaters.
A note and a question re the whistler’s mom ones. “What it did with the background painting” becomes clearer when you look at the original. What it did was, it conflated the framed painting above, with the gray floral drapery to the left. The spatial rhythm of the drapes is apparent. My question is, the original is wide, about 2×1 ratio. Whence the square ratio of the AI renders? Is that a built-in constraint of Midjourney?
All that said, the Stable Diffusion “Gothic” is just brilliant in so many ways. What did you say when that came out on the screen?
@4: Beautiful, but intense and possibly evil. Exotic fabric and clothing style. Probably a Bond villain.
@Madame X: No chains! I wonder if the AI was trained on the pre-censorship version of the portrait.
Did you ask the program to make Mona Lisa look 5-10 years younger?
Turns out if you boil our culture down to averages the creepiness becomes really obvious. The midjourny ones are better at least (lack of tits aside)
@Reginald Selkirk,9: did you click on the link in #1? Because that’s the inspiration for #4, and your description fits both.