![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
Garth Greenwell’s A Moral Education: In Praise of Filth starts off with this:
Greenwell is a college-level teacher of writing and English, and so are his students. But here are issues here that he touches on that, I think, deserve a closer look.
He says, “Within the small world of people who care about art…”, and we’re seeing that right now with the AI Illustration debate. The world of people who “care about art” is quite small, at least in the sense of ‘care’ that Greenwell applies to it: the people who want art to do well, to be both meaningful to the world and an exploration of how to make people feel something new and different.
Most of the illustrations I hang on my walls are nostalgic: movies, television shows, or concert posters that remind me of a story I enjoyed or an experience that I can only share second-hand with other people. And even then, the story is likely to have been simply affirming; to the extent that there was any moral to a story, it probably told me things I already believed, or things that I wish were true, such as “good always wins” or “love conquers all.”
I like controversial art; Omaha loves to twit me for my appreciation of some modern art and the explorations it made. All-white canvases where the artist has built up layers and layers of white paint in a way that shows different things if the light is held at different angles is a parlor trick I adore, but she finds it boring. But I rarely keep it in plain sight.
Most people, though, find art “boring” or “annoying.” To the extent that they notice art, it is in their face, challenging them to think differently. Mostly that reaction is “Aah, some asshole artist thinks he’s clever again.” They don’t care. They just want their world filled with prettiness, simplicity, understandability. Comprehension should be instantaneous; that you have to work for it is anathema to them. That you have to first build a context within which a work of art makes sense, that you should have to think, is anathema.
The “moralists,” the scolds who haunt school board meetings screaming that books be removed, books that educate children about the wider world, books whose topics include the idea that their teenager’s lives are equal to and independent of their own, that gay and lesbian and trans people (even children) exist and have a right to exist beside their own, that the culture they live in is imperfect and wounded by its past to an extent that harms women, children and minorities.
They care about “art,” but not in a good way. What they care about is what are is not. It’s not friendly. It’s not accessible. It’s not reaffirming. It challenges, it educates, it moralizes by showing the harms those screamers inflict on the world with their insistence that differences be hidden or killed. It tells kids that older people, yes even their parents, are hurting and killing other kids with their attitudes toward LGBTQIA+ people, Black people, and even women.
They want art to not be meaningful.
The state of the art in home AI Illustration is Automatic1111’s Stable Diffusion WebUI (SDW) distribution; installing it is a fairly technical undertaking but with plenty of on-line guides to help almost anyone who can navigate a spreadsheet should be able to get it up and running. And anyone who has a computer capable of running a modern video game such as Cyberpunk 2077, Doom Eternal, or Horizon Zero Dawn has the processor needed to run it.
Once up, you must acquire an SD model, of which there are literally hundreds. Once you’ve installed the model, you start the program and, in the app that’s presented to you, enter a prompt. A prompt could be “a cat wearing a space suit with a starry sky in the background,” “a beautiful girl wearing a dark blue Chinese dress glares at the viewer,” or “a tilt-shift photo of a tiny house on a leaf.”
“AI Illustration” has two processes running inside it: one generates a wildly random mosaic using the prompt (and parameters described below), and the other examines each mosaic and sends back data instructing the first on how to generate the next iteration so that it honors the prompt better.
Each SD model is a box, maximum 8 gigabytes in size, that has an uncountable number of images in it. Consider that, for each prompt, you can also set:
The free guidance scaling is a real number, not an integer. Assume people only ever use it to set “7.5,” halves. That means that for every unique prompt, not counting the resolution setting (which will cause wildly different outputs all on its own!), there are five trillion images.
There are five trillion cats in space suits. Five trillion pretty angry girls in a blue dress. Five trillion tilt-shift photos of a tiny house on a leaf.
On a store-bought “gaming” computer it takes somewhere between ten seconds and two minutes to generate a 640x480 image, given how many steps you ask for and which sampling algorithm you give it. Most of them aren’t very good, but they are recognizably close to what you typed in. A lot of them are repetitious, variations on the prompt in tiny, subtle ways, and most “AI Artists” spend their time filtering through all the bad ones to find the one or two that are convincingly lovely.
If you want hundreds of comforting images and you don’t care about what the art “says,” in fact if you’re offended by the idea that art should do anything other than affirm your already existing biases, then AI illustration is going to give you exactly what you want. Comforting, effortless, illustration that doesn’t clash with or chafe against the constraints of their drab, wretched lives.
One of the biggest complaints about AI Illustration is that, because it’s taking the billions of images artists and photographers have put onto the Internet since the Internet began, shattering those images into 8,589,934,592 (8 gigabytes) data points, and then remixing and rematching those data points into an almost infinite number of images, there’s no “there” there. The prompter didn’t put any meaning into the work; anything that seems to resonate with the viewer is entirely the viewer’s doing.
That’s pretty much what the viewer wants. They don’t want communication. They want comfort and beauty and reassurance. And AI illustration gives it to them.
So, sorry, Harjit, for most people, AI illustration is exactly what they want.
HERE’S A WAY of putting the problem: on one hand we want art to be free, and on the other we want it to mean. Not just to mean, but to be meaningful—to be useful for, and so maybe responsible to, other realms of life: our sense of community, say, or politics, our moral relations. … Within the small world of people who care about literature and art, the culture is as moralistic as it has ever been in my lifetime: witness our polemics about who has the right to what subject matter, our conviction that art has a duty to right representational wrongs.
Greenwell is a college-level teacher of writing and English, and so are his students. But here are issues here that he touches on that, I think, deserve a closer look.
He says, “Within the small world of people who care about art…”, and we’re seeing that right now with the AI Illustration debate. The world of people who “care about art” is quite small, at least in the sense of ‘care’ that Greenwell applies to it: the people who want art to do well, to be both meaningful to the world and an exploration of how to make people feel something new and different.
Most of the illustrations I hang on my walls are nostalgic: movies, television shows, or concert posters that remind me of a story I enjoyed or an experience that I can only share second-hand with other people. And even then, the story is likely to have been simply affirming; to the extent that there was any moral to a story, it probably told me things I already believed, or things that I wish were true, such as “good always wins” or “love conquers all.”
I like controversial art; Omaha loves to twit me for my appreciation of some modern art and the explorations it made. All-white canvases where the artist has built up layers and layers of white paint in a way that shows different things if the light is held at different angles is a parlor trick I adore, but she finds it boring. But I rarely keep it in plain sight.
Most people, though, find art “boring” or “annoying.” To the extent that they notice art, it is in their face, challenging them to think differently. Mostly that reaction is “Aah, some asshole artist thinks he’s clever again.” They don’t care. They just want their world filled with prettiness, simplicity, understandability. Comprehension should be instantaneous; that you have to work for it is anathema to them. That you have to first build a context within which a work of art makes sense, that you should have to think, is anathema.
When Pseudo-Moralists Attack
The “moralists,” the scolds who haunt school board meetings screaming that books be removed, books that educate children about the wider world, books whose topics include the idea that their teenager’s lives are equal to and independent of their own, that gay and lesbian and trans people (even children) exist and have a right to exist beside their own, that the culture they live in is imperfect and wounded by its past to an extent that harms women, children and minorities.
They care about “art,” but not in a good way. What they care about is what are is not. It’s not friendly. It’s not accessible. It’s not reaffirming. It challenges, it educates, it moralizes by showing the harms those screamers inflict on the world with their insistence that differences be hidden or killed. It tells kids that older people, yes even their parents, are hurting and killing other kids with their attitudes toward LGBTQIA+ people, Black people, and even women.
They want art to not be meaningful.
AI Illustration: A Brief Introduction
The state of the art in home AI Illustration is Automatic1111’s Stable Diffusion WebUI (SDW) distribution; installing it is a fairly technical undertaking but with plenty of on-line guides to help almost anyone who can navigate a spreadsheet should be able to get it up and running. And anyone who has a computer capable of running a modern video game such as Cyberpunk 2077, Doom Eternal, or Horizon Zero Dawn has the processor needed to run it.
Once up, you must acquire an SD model, of which there are literally hundreds. Once you’ve installed the model, you start the program and, in the app that’s presented to you, enter a prompt. A prompt could be “a cat wearing a space suit with a starry sky in the background,” “a beautiful girl wearing a dark blue Chinese dress glares at the viewer,” or “a tilt-shift photo of a tiny house on a leaf.”
“AI Illustration” has two processes running inside it: one generates a wildly random mosaic using the prompt (and parameters described below), and the other examines each mosaic and sends back data instructing the first on how to generate the next iteration so that it honors the prompt better.
Each SD model is a box, maximum 8 gigabytes in size, that has an uncountable number of images in it. Consider that, for each prompt, you can also set:
- The size of the image: 512x512, 640x480, even 1920x1440 if you have a very expensive graphics card.
- Free guidance scaling: a number between 1 and 30 that determines how much of the prompt the two AIs will use to generate and affirm that the image adheres to the prompt.
- The iterative steps the two AIs will use to generate and affirm the way the image being generated adheres to the prompt. More iterations mean a better look, but it also means it takes more time to generate.
- The sampling algorithm used to make that determination. There are currently 20.
- A starting point in the box for where the generative AI should grab the tiny tiles for the mosaics with which it works. There are 4,294,967,296 valid starting points.
The free guidance scaling is a real number, not an integer. Assume people only ever use it to set “7.5,” halves. That means that for every unique prompt, not counting the resolution setting (which will cause wildly different outputs all on its own!), there are five trillion images.
There are five trillion cats in space suits. Five trillion pretty angry girls in a blue dress. Five trillion tilt-shift photos of a tiny house on a leaf.
AI Illustration: Giving the People What They’ve Always Wanted
you cannot convince me that we as a human species, who carved art on cave walls between hunts, who penned poems during war, that made anything and everything from the photos we take to the buildings we make into art forms, would rather have AI make our art for us. you just can't
— harjit is trying to write (@_harjitsingh) January 3, 2023
On a store-bought “gaming” computer it takes somewhere between ten seconds and two minutes to generate a 640x480 image, given how many steps you ask for and which sampling algorithm you give it. Most of them aren’t very good, but they are recognizably close to what you typed in. A lot of them are repetitious, variations on the prompt in tiny, subtle ways, and most “AI Artists” spend their time filtering through all the bad ones to find the one or two that are convincingly lovely.
If you want hundreds of comforting images and you don’t care about what the art “says,” in fact if you’re offended by the idea that art should do anything other than affirm your already existing biases, then AI illustration is going to give you exactly what you want. Comforting, effortless, illustration that doesn’t clash with or chafe against the constraints of their drab, wretched lives.
“But It Doesn’t Mean Anything!”
One of the biggest complaints about AI Illustration is that, because it’s taking the billions of images artists and photographers have put onto the Internet since the Internet began, shattering those images into 8,589,934,592 (8 gigabytes) data points, and then remixing and rematching those data points into an almost infinite number of images, there’s no “there” there. The prompter didn’t put any meaning into the work; anything that seems to resonate with the viewer is entirely the viewer’s doing.
That’s pretty much what the viewer wants. They don’t want communication. They want comfort and beauty and reassurance. And AI illustration gives it to them.
So, sorry, Harjit, for most people, AI illustration is exactly what they want.