DALL-E 2 is artificial intelligence software that can turn anything you type into art, in any style. You want a portrait of a panda in the style of Renoir? Here ya’ go!
A Sugarplum Fairy eating a cheeseburger in the style of Kehinde Wiley? Try this! Said one woman, “That is frightening and fascinating all at the same time!”
People are using DALL-E to make music videos…
… as well as children’s books and magazine covers.
I’ve even used it to illustrate “Sunday Morning” stories.
DALL-E 2 and its rivals, like Midjourney and Stable Diffusion, are available to anyone; they’re inexpensive, or even free, to use. It’s easy to see how this technology will change the game in graphic design, interior design, architecture, fashion, and moviemaking.
DALL-E’s creator, Aditya Ramesh, works at OpenAI, a company founded in 2015 by Elon Musk and others. “The goal of the company is to develop artificial general intelligence,” he said. “And by that, we mean an AI that can do all of the things a human can, and to deploy it in a way that’s safe and maximizes the positive benefits to society.”
To train DALL-E (the name is a mashup of Dali, the artist, and Wall-E, the Pixar robot), the company fed it 600 million labeled pictures from the Internet.
“It’s not merely cutting-and-pasting together; its understanding of images is sort of more conceptual and abstract, sort of like how a human would use inspiration from all of the images that he or she may have seen in their lifetime,” Ramesh said.
I know what you’re thinking: This is going to put a lot of artists out of work.
Meet concept artist Karla Ortiz, who has designed characters, creatures and costumes for many Marvel movies, like “Doctor Strange.” “Why would someone hire someone, when they could just get something [AI] that’s ‘good enough’?” she said.
But her biggest concern isn’t unemployment; it’s that professional artists’ work has wound up in OpenAI’s database. That’s how DALL-E knows how to imitate the style of Norman Rockwell, or Picasso, or Ansel Adams, or living, working artists.
“It’s an invasion of privacy,” said Ortiz. “It’s an invasion of our ability to consent to be into these, you know, data sets. Nobody asked us. The way to fix this is to do so by building data sets that are completely full of public domain works, and then, any further kind of expansion from that, done so by licensing agreements.”
The idea being, if an artist chooses to “opt in” to the AI’s database of images, and become a part of its algorithms, the artist would be compensated.
The AI art companies say they’re listening. For example, Stability AI recently announced that it will let artists “opt out” of future versions of its database.
But OpenAI also worried about other downsides, like people churning out AI-generated images containing porn, violence or misinformation. Ramesh said, “When we trained the model, we filtered out images of weapons, blood, gore.”
So, if someone entered the phrase “The president killing kittens” into DALL-E’s image generator, it would give an error message in return. “It won’t let you do that,” Ramesh said.
DALL-E also tries to compensate for the racial and gender stereotypes in the internet’s universe of pictures. So, even though 90% of the doctor pictures on the web may be White men, Ramesh said, DALL-E would try to even things out.
But not all AI companies have built in those kinds of safeguards. According to Emad Mostaque, the CEO of Stability AI, “There deserves to be lots of different views and lots of different perspectives on this. And as a society, we have to come together and figure out what the best way to use this amazing technology is.”
The program Stable Diffusion, from Stability AI, is open source, meaning it’s free to anyone, without restrictions or guardrails. That approach has rung a lot of alarm bells.
Mostaque said, “We think putting this out to the open so people can see the power of the technology and then figure out together how we can mitigate the harms, is superior to it being in the province of unelected companies.”
Some Stable Diffusion fans do produce harmful and shocking images, but they rarely see the light of day, according to Mostaque, because Twitter and Facebook screen them out. “If you put it onto social media, or you put it out there, it’s treated just like any other bad content,” he said.
The state of the art in AI art is getting there. These early apps still have trouble with text, faces, and generating the usual number of fingers.
But they’re improving fast. And meanwhile, AI apps that generate audio and video are already in testing.
For artist Karla Ortiz, these are distressing developments. She believes that there’s value in the creative process itself: “It’s therapeutic. It’s inspiring. It’s communicating between one human and another. AI tools can’t do that just yet.”
But Stability AI CEO Emad Mostaque is all-in on the AI arts. “I think it’s one of the biggest leaps forward we’ve had in technology since maybe the internet,” he said. “It’ll create brand-new industries, and it will make media even more exciting and entertaining. I think that creates loads of new jobs.
“It’s coming, inevitably. And I think it’s just going to change just about everything.”
For more info:
Story produced by Sara Kugel. Editor: George Pozderec.