Projects like these drawing programs commonly use a two-part process called GAN, for “generative adversarial network.” First, you give the computer data to look at – millions of pictures of things you want it to be able to draw, which in the case of these programs is All The Things. Then part A of his logic starts drawing things, and part B looks at them and gives feedback on how close the pictures are to what they’re supposed to be. Ultimately, humans look at the output and decide what the machine needs to see more examples in order to improve.
Here is an article of the “parents” of such a program, talking more about what they do and what they can and cannot offer. They make the case for their baby, named AICAN, but near the end they offer a helpful observation:
Yet there is something missing in AICAN’s artistic process: the algorithm can create attractive images, but it lives in an isolated creative space that lacks social context. Human artists, on the other hand, are inspired by people, places and politics. They create art to tell stories and make sense of the world.
I’m going to put two videos here, each under 15 minutes long, which you don’t need to see to enjoy the footage below, but they do give a bit more information about what’s going on.
This video is a very good explanation of how it works and what happens in the machine (the detail starts at around 6 minutes). I would recommend this one if you want to learn more. [13:32]
This one is specifically a guide to Midjourney, the AI that created almost all of the lyric art for the songs we’ve seen in the past week. It’s more focused on how to use that specific product, but since we’ve seen a lot of Midjourney’s production, I thought some people would like to know how it’s used. This is not an endorsement on my part of the attitudes towards art expressed here. [12:38]
So as you know I went to pencil.com and started toying with the phrase “kitchen table kibitzing”. I started by just entering “kibitzing kitchen table” to see what would happen. I made three passes to this, then, at the site’s suggestion, I tried adding “high definition”. What I learned was that the images the AI had been training on didn’t offer it any knowledge of this “kibitzing” I was talking about. You can see it took a wild guess and added lots of what appears to be “kibble”, if that’s what I meant. This path was not likely to become more interesting anytime soon, even if the warped utensils and the pi-shaped bagel were a bit trippy. (Links throughout this text are to original size images.)
So there, on the site’s further advice, I tried “kibitzing kitchen table paint”. NOW we were getting somewhere. He was still making pictures of strange scenes, with deformed people if there were any, but they were surreal in a much more interesting way. The two enlarged frames were my favorite of each batch. Special mention to the sparsely legged table with the platter of giant garlic-pistachio hybrids and tiny, shall I say, guillotine?, and the MC Escher chair. [Other one: cocktail swizzle sticks gather at the table.]
Oh, and seeing this chair, I tried MC Escher – the results were a bit disturbing. I feel like Escher’s work is really not understood like some artists. His images are so symmetrical, and this image seems almost organic in its twisted mess. It’s like a monster that ate an Escher.
Before setting off on the path “painted by”, I took a small detour inspired by the Bayeux Tapestry KTK title picture. I started with “tapestry”, but it turned out to be more Carole King-y, so I added “medieval”. (Okay, those specific links don’t refer to the full size image, but this one is.)
I’ve put all the images in sewn together bars grouped in very rough chronological order (emulated artists), as there are so many, and logs aren’t very user friendly to place lots of individual images to the left and to the right of the text. Also, I’m moving the large image links to the captions from here to the bottom.
I like that he decided on his own to depict frames around images and/or white space as if they were hanging on a wall and we were a little out of the way. As for the image content, it looks almost normal at this scale, but if you look at the enlargements, they get weird. This “Degas” for example – it’s really not clear which elements of the picture are dancers and which are furniture.
I opted for “a pointillist” because he knows only one work of Seurat’s obvious choice. This first image, of what appears to be a robed shrub sitting at a table by the sea, is more interesting than nine tries at the island of La Grande Jatte. On the other hand, he only knows one work by Klimt and gave me nine. They all pretty much look like the actual work; this one went the furthest by turning the couple into a tall, chic teapot with a black knob on top.
As discussed, he particularly hurts realistic humans, since he doesn’t know what they are or why it would bother us when they are distorted in ways not compatible with life. The image of “Dorothea Lange” is frightening. The figure of “Diego Rivera” in his flippant fried egg hat is just a “painting”, so it’s not disturbing.
There’s a girl in a real Carl Larsson painting wearing a dress like this, but today the dress arrived at the kibitz with no occupant. The “Ansel Adams” kitchen table in the desert sounds like some sort of Second Amendment statement, doesn’t it?
Once we get to a time where there is art that isn’t figurative to begin with, the machine renditions get a whole lot better. I’ve loved every one of the “Joseph Cornell” boxes she’s made, and the “Calder” kitchen table is awesome.
Oh! I see I still have something hidden in my working draft that I wanted to show you. Wonkette has sometimes used AI art as title images on their stories. They use an AI called DreamStudio, and they still use it to create snark, because they’re Wonkette. I enjoyed the art for ther recent story by Doktor Zoom; the image is captioned, “What the AI thinks Ron DeSantis and an AR-15 look like. I removed the extra hand it gave DeSantis.
Don McLean: vincent [4:06]
Paul Simon, with yMusic chamber music quintet: René and Georgette Magritte with their dog after the war [5:07]