Inside The World of Uncanny AI Twitter Art

The only thing more surreal than a Salvador Dali painting is artificial intelligence’s conception of a Salvador Dali painting. Or at least, that’s how it appears, in “The Persistence of Memory by Lisa Frank,” a piece of art created using AI software by Sam King, a philosophy major Northwestern University, who creates and curates AI art on his popular Twitter account @images_ai.

The image is like a technicolor Rorschach in composition, with bursts of Frank’s signature rainbow gradient mixed with Dali’s distorted objects. The effect hits a particular pang in the valley of the uncanny, like an alien’s conception of a painting that’s as mesmerizing as it is uncomfortable.

King first discovered AI art on Twitter last year, after coming across a series of wild images posted by the founder of his favorite vaporwave label DMT Tapes FL. He started poking around and realized there were a number of people creating images using machine learning, so he decided to curate all the art in one place. Now @aitwitterimages has more than 47,000 followers.

At a very basic level, AI art is created like this: Using open-source software, you type a string of words into a machine that has a library of thousands of images, and the machine spits out its conception of those words in an image. For example, the phrase "Could any amount of love ever be worth the pain of losing it?" spat out an abstract expressionist image of a lone cart on a farm at sunset, conveying a tone of melancholic loneliness. The phrase, “this emotiguy does not exist” gives you petrifying emojis that look like a warped off-brand version of Picasso’s cubism era.

“I think it’s really fun that you can give [the machine] something that you might imagine is very vague or that you might imagine is too flowery for a machine to understand and it will still capture the essence of what you were feeling when you read the statement,” King tells NYLON. “I think that’s remarkable that you can give it a line of poetry and it will encapsulate the feeling of the poem, despite the poem using all sorts of simile and metaphor and analogy in its writing.”

Like most things on the internet, people have opinions about AI art – namely, that it isn’t art. But it has an alien aesthetic of its own, one that could only be created by a human using a machine, and one that is beautiful precisely because of its randomness, not in spite of it. For example, the machine doesn’t do borders well, nor does it understand object permanence.

“It doesn’t understand three-dimensional space in the way that we do. When it creates something like a sense of depth, it’s coming from just having seen images from 2D images, so it just doesn't construct objects the way we understand objects,” King explains.

You could be quick to dismiss AI twitter art as its own strange internet trend — but King believes AI art is going to be yet another tool in an artist's toolbox. The latest iteration of Photoshop even has something called neural filters which use AI technology; one of the tools takes old black and white images and colorizes them.

“I think the primary application of this in the future is as tools we employ in tandem with more traditional forms of art,” King says. “I promise it is much much easier to do than it looks. I think that more people should have an understanding of it – I think it is going to play a major part in the future of all art creation.”

NYLON spoke with King about how to make AI art, the future of digital art, and about why everyone went crazy over his Salvador Dali-Lisa Frank mashup.

How did you start making this kind of art?

I’m a student at a university in northern Illinois and I follow this the DMT tapes FL on my personal Twitter, and in late spring of 2021, I saw the guy who runs that label was posting these crazy images. I looked into it and found out how to do it and discovered the software, which was free to use, open-source, and really user-friendly – I don’t have any coding experience really; I’m a programmer. I saw that nobody was really following the people who made the art, and the artist who created the software didn't have much of a following, so I figured it might be a good idea to create an account specifically dedicated to creating stuff with machine learning and promoting artwork created with machine learning. Now we're at like 47,000 followers in under a year, which is nonsense.

So you were the first Twitter account to curate AI art in one place?

Yes, that seems to be the case, at least initially. Obviously it's been massively expanded since then, so it's in many different corners. There were many many people doing art with artificial intelligence before; I feel like I’m slightly responsible for the specific software [VQGAN + CLIP by Katherine Crowson] I was using getting really big because I wrote a tutorial for it, put it in a Google doc and it’s been my pinned tweet since June of last year and it’s been used by thousands of people at this point, so a lot of people have learned to use this specific technology through my account.

Can you give us a high-level explanation of how you make these images?

The software takes a sentence you write and it turns it into an image. So you put it in the machine and you post the raw output, and there’s a fun novelty there in getting to see the image the text gives you. Last year, there was a whole barrage of messages people were sending me with all sorts of keywords and a lot of people asked for similar things. For a while, I was just posting people’s requests and messing around in the software, adding additional pieces to the text for my own amusement to see if I could create particularly cool images. Sometimes I was thinking of prompts myself. I wasn't trying to be beholden to the replies, but people were constantly suggesting things I thought would be interesting to see realized, so I was just going along with it.

Now, more recently I’ll discover a new piece of software and I will initially try to make a bunch of images that will test the limits of the thing, so I can get to grips with it personally. Then I will start to use it on my own terms as just another tool in my skill set when it comes to creating digital art. [The AI image software] Diffusion or Photoshop are just additional tools in the roster now. I create composite mixed media images and I know if there's a particular look or style I'm going for that involves an AI-generated feel, then I know I can go to [the software] VQGAN and get this particular effect or Diffusion and get this particular effect. And if the piece of art I’ve created happens to include machine learning, I will post it to the Twitter account.

What kinds of phrases were people suggesting? Were there any that stood out or came up a lot?

Memes would come up all the time. People would look for joke stuff. They’d ask about Shrek all the time. They’d ask for biblically-accurate angels, things like that. Personally, I really like it when people give lines from poetry. There were a few times where I’d ask for people to give me lines from poetry. There were a few times where I would make a series of images where it was like a different picture for each line of a poem, because I think it’s fun when you give the machines more esoteric or out there, abstract language rather than ask for something hyper-specific.

Obviously if you give these machines very hyper-specific requests, you’ll get something you’re more or less expecting. And it will take you by surprise to an extent, but I think it’s really fun that you can give it something that you might imagine is very vague or that you might imagine is too flowery for a machine to understand and it will still capture the essence of what you were feeling when you read the statement. I think that’s remarkable that you can give it a line of poetry and it will encapsulate the feeling of the poem, despite the poem using all sorts of simile and metaphor and analogy in its writing. I think that’s fantastic.

Where are these images sourced from? Where does the machine pull the images it uses to create these pictures?

I think it depends on which one. The [software] VQGAN is trained on a few different models. They’re all really big image data sets. The full one is imagenet.org and they’re like an open source collective that puts this big set of images together. Wikiart is another big one. Some people will have their own specific fine-tuned models that people will post. I’m not involved in the technical aspect of putting the software together.

“It’s a tool you can use the way you’ve used a paintbrush, the way you’ve used knives or glue or clay. It’s material – a substance you can create with.”

What are people’s criticisms of AI art? Do they say it's deritivite? It’s not art? What’s the pushback?

All the time. It’s interesting the kind of people who will make those arguments, because sometimes it’s like high art philistines, the typical group you’d expect to say, “This is art, this is not art.” Sometimes it’s people in disciplines who feel threatened by it. There’s a lot of pushback from photographers for example, because I think there's a feeling among some people that AI art is trying to replace them and trying to replace the schools of art they work in. They’ll come along and say, “you just have to type in a few things and then you run it and it’s so easy and it's so lazy, so it’s bad art.” I understand that to an extent, but I think firstly, they don’t realize the extent to which it’s a discipline in itself. In the immediate present, creating something is really quick, but getting to grips with the software and understanding the nuances of what it will do with various prompts is an art form that requires time and labor in order to understand it and make good use of it. That’s something that often people don’t consider.

The other argument is that it’s not human-made art, because a machine’s doing it, so you as an artist aren’t legitimate because you’re outsourcing everything to the machine. I think different people are going to have different perspectives on that depending on how they conceptualize the machine itself. I imagine these as tools and to an extent maybe collaborators, to the extent that I’m working with it, but I think often initially it feel like I’m working with these as peers, as collaborators and I’m exploring the space with them and overtime they develop a feeling of: you’ve understood it, you’ve mapped it out and now it’s a tool you can use the way you’ve used a paintbrush, the way you’ve used knives or glue or clay. It’s material – a substance you can create with. It does seem to evoke some passionate responses from completely different sides. Some people think it’s really beautiful and some people are freaked out beyond words. It's really interesting.

I love your Dali-Lisa Frank-inspired piece, The Persistence of Memory by Lisa Frank. That one is so funny to me.

That’s the first image that blew up. That’s where 20 to 30-thousand of the followers came from, from just that image.

Why do you think that one resonated?

For a few reasons. It’s a vaporwave reference and one big song from that genre is a song that has that in the title. I think it’s in in Japanese, so someone tweeted that at me and I thought okay, that’ll be funny and I think I started with the original painting by Salvador Dali as the starting image and I think I left it running overnight, so it ran for like 8000 iterations and became the image it became. It Lisa Frankified the whole thing. I live in America, but i’m not American, so I was not familiar with Lisa Frank as an artist when I made the image and then I posted it and instantly everyone was like, I love Lisa Frank. I looked it up and perhaps it’s that a lot of people have intense memories of Lisa Frank, that seems like a strong theme in replies to it: that I can’t believe an AI made this and I can't believe it captured her style so well.

Is this the future of digital art? Is it a passing trend or is it too soon to know?

The guy who kicked all of this off was @advadnoun and he created a combination of [the software] CliIp with GAN and that was the notebook that inspired the [software] VQGAN+CLIP that I started using and he’s been a follower for a long time. We're kind of buddies now and he works for Adobe. He works on implementing some of this software into Photoshop, so in the most recent edition of Photoshop there are these things called neural filters which is essentially an application of artificial intelligence for various things in Photoshop. One of the tools is a colorizer for black and white images. It works amazingly well. You give it a picture from the 1940s of people standing around and it will colorize it remarkably well. It’s not perfect yet, but it will be within a year.

I think the primary application of this in the future is as tools we employ in tandem with more traditional forms of art. I think there was a phase for a while of people posting raw output of machines which I did help to popularize. I have moved away from that. I don’t do that as much anymore. I've started to post a lot more mixed media and integration of this style of art with other more traditional styles of art or incorporating multiple different forms of AI in the same image. It's just going to be the normalization of AI tools in the artist's roster. It’s not gonna replace anyone. That’s a common worry: that this is just going to become the way that art is created and I think that that’s just ridiculous. It seems overwhelming and daunting if you don’t understand the terminology, but I promise it is much much easier to do than it looks. I think that more people should have an understanding of it. I think it is going to play a major part in the future of all art creation. In the past few days I've gotten into an AI machine that creates music. I think this will pervade all forms of art.

People have misconceptions about artificial intelligence and machine learning and the way it works and I won’t pretend to understand the intricacies of the tech itself, but as someone who understands the way it can be applied and who has an understanding of the current scene in its application, it's well worth getting into. There’s really really interesting stuff going on and it’s worth participating in and also it’s free it’s completely free to use. It’s all free and open source and you can modify it if you’re not a programmer and if you aren’t a programmer, if you don’t have any money, you don’t need to spend anything to make really interesting art. I think that’s so important. I'm really big on accessibility of it.