Someone in the “AI Art Universe” Facebook group called it “art harvesting.” It’s an interesting analogy: sprouts planted by many other people are ‘scraped’ into a giant blender that sorts and readies them to be grown into exotic new gardens. But it’s more than a poetic analogy— it’s a worldwide phenomenon, way bigger than a garden. It’s a jungle of fields and plantations, meadows and forests filled with fantasy characters and creatures, scenes and settings that could be in the distant past, the far future, or another galaxy. And it’s springing up, morphing, regenerating before our eyes. Some of the results are dark and ugly, some are eerily beautiful, and all you have to do to participate is type a prompt that describes your vision. A minute or so later, a suite of images springs up on your screen, ready to be enhanced by you (and, apparently, by anyone else).
I trolled around for a while, trying to find an AI-generated garden “good enough” to show as an example. I finally settled on an alien landscape credited to Bryan Price on NightCafe.studio. With it came a 25%-off-my-first-month offer. Ah yes, another income-generator for someone who is not me, i.e. for NightCafé and all the similar sites that are popping up.
Traditional illustrators are up in arms. On Thursday, December 22, the Society of Illustrators posted this message on Facebook, Instagram and Twitter.
In less than 24 hours, this collaborative post by the award winning duo of Society of Illustrators President Tim O’Brien and illustrator Edel Rodriguez had more than 8,500 likes on Instagram, 16.2k likes, and 3,280 reposts on Twitter.
For more than 30 years, O’Brien has been painting meticulous oil portraits of famous people, from his hero Muhammed Ali to Elon Musk, and many have been featured on the cover of TIME. He and his colleagues are insisting that illustrations for publication must be created by real, thinking humans who interact with real clients and use real artists’ tools. “The sudden availability of artificially designed images creates a moral challenge to the illustration community and to the broader design community,” O’Brien said. “We are at the critical point at which illustrators and designers must value human interaction and reject the output of AI image generators. The inclusion of a credit highlighting an AI generator should bring on a sense of shame.”
O’Brien went on to note that athletes are subject to drug tests for trying to enhance their performance artificially, and those who fail are punished. “Humans can run faster, jump higher, and perform better using synthetic means, [but] we as humans are interested in what a human alone can do. That’s what makes us human.”
Illustrator Victor Juhasz, best known for caricatures that have graced the pages of Rolling Stone, TIME, Newsweek, and many other publications, takes the argument a big step farther. “The current craze for AI-generated ‘art’ is a symptom of a disease,” he said.
Juhasz did not mince words. “The temptation to take the fast, easy way rather than put in hard work is enormous. Contemporary society thrives on celebrity, fame and notoriety, and much of it has nothing to do with honest craftsmanship. It’s about the con and getting away with it.”
Other notable illustrators like Anita Kunz, known for her New Yorker covers and feminist responses to classic art have spoken out on how much they hate seeing their work scraped into databases. Karla Ortiz, a painter, printmaker, and concept artist at Marvel Studios, has been especially vocal on social media, posting impassioned arguments against the commercial use of AI art and spearheading a GoFundMe campaign to hire a lobbyist to make the voices of artists heard.
At the present moment, the creative heads of magazines sound largely uninterested in AI. Michael Mrak, the creative director of Scientific American, a science publication with over 10 million subscribers, “[sees] no reason to replace real artists with AI-generated anything.”
“AI can generate interesting and elaborate imagery, but there are many problems from a legal and moral point of view,” he continued. “AI-generated art cannot be copyrighted and therefore has potential legal issues attached to it, a principal one being that it uses art from across the internet to make the final image. That, and the fact that it scraped or pulled copyrighted art into its learning algorithm.”
Art director and designer Alexander Isley treasures his one-on-one collaborations with artists. “I have never used AI-generated artwork, and have no interest in doing so, unless it’s in the context of how odious it is,” he said. “From all I’ve seen and read, machine-generated artwork is based on modifying, remixing, or adding to real artists’ existing work without acknowledgment or compensation. With commissioned artwork, sketches and revisions are often required. How does this process work with AI-generated images? I can’t deny that the results can be interesting to look at, but it’s a fun parlor trick.”
While art directors might not see AI as a threat, the competitive aspect of design complicates matters. Will AI-generated art be eligible to win contests and grants?
“The short answer is yes,” was the initial answer from Patrick Coyne, editor/designer of Communication Arts, one of the world’s most important design publications. “We always tell jurors to select work based on the quality of the idea and the execution. We celebrate compelling imagery regardless of how it was created.”
Managing Editor Michael Coyne noted that Communication Arts had already featured a few campaigns that used AI-generated art “because they were interesting or appropriate applications for AI as an artistic tool rather than a medium on its own.” He cited an ad campaign by Dentsu Creative Portugal for Jardim Sonoro, an electronic music festival held in a national park near Lisbon, is an example. According to the agency’s creative directors, the challenge was to blend the musicians’ portraits with natural elements. “We learned that AI is a great tool,” they commented. “We are still at its beginning and will certainly see significant developments that will dazzle us all. But it won’t replace anyone. It needs someone to guide the creative process.” They added, “Novelty and discomfort often lead to great work.”
However, a few days later, the Communication Arts‘ team’s stance evolved. “We’ve been approached by several illustrators upset over our position regarding accepting entries for the Illustration Annual produced with text-to-image AI software,” Patrick Coyne wrote via email. “While I still see the long-term potential for AI-assisted creative exploration, I better understand the position that illustrators and photographers are currently facing with copyright infringement and the unauthorized use of their work to ‘train’ the current crop of text-to-image AI software. Consequently, we are reversing our position and will not be accepting AI text-to-image generated submissions in our Illustration competition.”
Hobbyists have a different relationship to the software. Daniel Rocha of São Paulo is an active contributor to Facebook’s “AI Art Universe” group, and one of the many thousands of people who make AI art for fun. “I use [Mage’s Standard Diffusion program] daily, many times a day,” he said. “I click ‘enter’ on a prompt many, many times, until I get something good or see that I need to change it because something is not nice. I’ve generated more than 22,000 pictures, but that’s not at all time-consuming, since all I have to do is click, click, click.”
Oddly enough, Rocha works in Brazil’s patent and trademark office, where he analyzes the registrability of trademarks. However, “that has nothing to do with what I do on Mage,” he clarified. “I think it is an extremely useful tool for artists. They can use it to fill in details or compose a complex scene extremely fast.”
Stable Diffusion can be trained to fit an author’s style, which allows them to make grandiose scenes in a short time, in their own style. “An amateur like me can reproduce the work of a skilled artist, art that could surpass in quality and inspiration the Sistine Chapel ceiling,” Rocha continued. “That took years for Michelangelo to make, and [similar work] can now be completed in a few days or weeks. Right now, the artists are too scared, but I think they will come around soon.”
Since I have family members who like to play with DALL-E, we decided to try it ourselves. I went in wondering if I could re-create one of the world’s most iconic posters, Milton Glaser’s 1966 “Dylan.” When I used Mage, the results were dismal. Apparently, the Mage database doesn’t have the stuff. We had no luck on DALL-E either (“does not follow our content policy”), but got meh results with Midjourney, where we typed “/imagine the famous 1966 Milton Glaser Bob Dylan poster” and got:
The curly hair must have gotten scraped in, along with some old album covers. And possibly black-and-white portraits to which the photographer owns the copyright. Then we tried: “/imagine the famous 1966 Milton Glaser Bob Dylan poster, but for Lady Gaga” and got:
The whole process took about three minutes. Fortunately— for now, at least— AI isn’t giving Milton Glaser’s brilliant work any serious competition.
To get clarity on where AI stands in regards to legality, I reached out to Martin Schwimmer, a partner at top-rated New York intellectual property law firm LeasonEllis. In his opinion, text-to-image AI models “present novel [new, unexplored] legal issues, including the extent to which the creator of the repository of images makes use of images that were previously displayed on the internet, and to what extent can an AI model look at an image and derive ‘rules’ about that image.” While that language is a little murky to me, it sounds like the lawyers are working on it.
However, Schwimmer didn’t agree that all AI repositories consist of ‘scraped’ images without regard to copyright. For example, he said, a repository named Laion consists not of images, but links to images, which apparently makes a legal difference.
As to who owns the so-called final product, Schwimmer said that he views AI models as one more tool that helps users generate content. “The copyright analysis will be comparable to the analysis we use today when artists use the various illustration tools, graphics editors, paint programs, and other digital art tools: Is the work sufficiently original when divorced from the accompanying tools?”
For now, that will be the last word.