OCTOBER 16, 2022
- OpenAI, a company founded by Elon Musk, just made its DALL-E image generator open to the public.
- Artists say they work for years on their portfolios and people can now make copycat images in seconds.
- But some AI companies argue that the new artworks are unique and can be copyrighted.
Greg Rutkowski is an artist with a distinctive style: He’s known for creating fantasy scenes of dragons and epic battles that fantasy games like Dungeons and Dragons have used.
He said it used to be “really rare to see a similar style to mine on the internet.”
Yet if you search for his name on Twitter, you’ll see plenty of images in his exact style — that he didn’t make.
Rutkowski has become one of the most popular names in AI art, despite never having used the technology himself.
People are creating thousands of artworks that look like his using programs called AI-image generators, which use artificial intelligence to create original artwork in minutes or even seconds after a user types in a few words as directions.
Rutkowski’s name has been used to generate around 93,000 AI images on one image generator, Stable Diffusion — making him a far more popular search term than Picasso, Leonardo Da Vinci, and Vincent van Gogh in the program.
“I feel like something’s happening that I can’t control,” Rutkowski, who is based in Poland, told Insider. “My name is being used a lot to generate AI images, along with the names of other working artists.”
“Dragon Cage” by Greg Rutkowski. Greg Rutkowski
AI-image generators create images that are unique, rather than collages pulled from stock images.
Images created when Insider typed “Dragon battle with a man at night in the style of Greg Rutkowski” into Stable Diffusion. Stable Diffusion
A user simply types words describing what they’d like to see, referred to as “prompts,” into a search bar. It’s a bit like searching Google Images, except the results are brand-new artworks created using the text in the user’s search terms as instructions.
One of the most common prompts is to use the name of an artist to create something mimicking their style.
“People are pretending to be me,” Rutkowski said. “I’m very concerned about it; it seems unethical.”
Simon Stålenhag, an artist and designer based in Sweden, told Insider that although he isn’t against AI-generated art in principle, he does take issue with how some people are using the new technology.
“People are selling prints made by AI that have my name in the title,” he said. “Something like — ‘Rusty Robot in a field in the style of Simon Stålenhag’ — which is a super aggressive way of using this technology.”
He’s seen people be hostile when they share an AI image in his style on social media. “People have tagged me and said that they’re gonna make me lose my job or something like that, they’re really harsh and aggressive,” he said.
He believes AI-image generators are “not in the hands of artists right now. It’s in the hands of early adopters of tech.”
Rutkowski, who uses both digital tools and classic oil on canvas for his work, is worried that this explosion in imitation art means his style — which has seen him land deals with Sony and Ubisoft — might lose its value.
“We work for years on our portfolio,” Rutkowski said. “Now suddenly someone can produce tons of images with these generators and sign them with our name.”
“The generators are being commercialized right now, so you don’t know exactly what the final output will be of your name being used over the years,” he said.
“Maybe you and your style will be excluded from the industry because there’ll be so many artworks in that style that yours won’t be interesting anymore.”
An explosion in imitation
More and more consumers are using AI-image generators.
OpenAI, which Elon Musk co-founded in 2015, made its DALL-E image generator open to the public in September. Before lifting the waitlist, OpenAI said the program already had more than 1.5 million users.
Liz DiFiore, the president of The Graphic Artist Guild, an organization that supports designers, illustrators, and photographers across the US, said the ease with which AI can copy styles could cause financial fallout for artists.
“Artists spend a lot of time throughout their career, and make a lot of income, on being able to license their images and being sought after specifically for their style,” she said.
“So if an AI is copying an artist’s style and a company can just get an image generated that’s similar to a popular artist’s style without actually going to artists to pay them for that work, that could become an issue.”
US copyright law only protects artists against the reproduction of their actual artworks — not from someone else mimicking their style.
Some of the most popular AI-image generators — which include DALL-E, Midjourney, and Stable Diffusion — have policies in place to prevent consumers from using their products in certain ways. OpenAI, for example, prohibits the use of images of celebrities or politicians.
All three programs block users from creating “harmful content” by filtering things like nudity and gore.
Insider asked representatives from DALL-E, Midjourney, and Stable Diffusion if they have any measures in place to prevent images being created that mimic the style of working artists.
A representative for Stable Diffusion said the company was working on an opt-out system for artists who don’t want AI programs to be trained on their work.
The spokesperson added that an artist’s name “is only one component of a diverse set of instructions to the AI model that creates a unique style that is different from an individual artist’s style.”
Representatives for Open AI did not specify any measures in place to protect living artists but said the company would seek artists’ perspectives as it expanded access to DALL-E.
Midjourney didn’t respond to Insider’s questions.
AI data training
AI-image generators “train” by learning from large sets of images and captions. Representatives from Open AI said both publicly available sources and images licensed by the company make up DALL-E’s training data.
Representatives for Stable Diffusion said the program uses web crawls to gather information and images.
Rutkowski thinks living artists should have been excluded from the databases that train the generators.
“I’m not against the AI overall, I think it’s a good technology. But I think they should have excluded artists’ names from the program,” he said.
Another designer and illustrator, RJ Palmer, dubbed the generators actively “anti-artist” on Twitter because he said they are “explicitly trained on current working artists.”
Artists can check if their work has been used to train AI programs on a website called Have I Been Trained, which the German artist Mat Dryhurst and the American sound artist Holly Herndon created.
The pair have been working on tools to help artists opt-out of AI data-training sets. The website filters through around 5.8 billion images that are in the dataset Stable Diffusion and Midjourney use to train their programs.
Other artists feel they should have been asked for consent for their images to be scraped for the data used to train AI generators.
Stålenhag said it would have been nice to be asked if he could be included in the training data, but said it was an inevitable consequence of putting art on the internet.
“I see it as being very similar to how artists already work,” he told Insider.
“We do copy other people’s ideas and styles and designs, and we take stuff,” he said, noting that he also doesn’t think AI art is good enough quality to be a “threat” at present.
“There’s a hype around AI that I think is weird because I just don’t think that it’s very good,” he said. “I don’t see it as a threat because the visuals are not as good as what artists can create.”
Copyright laws around AI images are murky
It’s unclear whether copyright laws will protect the new artwork that AI programs generate.
“Copyright issues around AI is probably one of the biggest areas that we are focused on,” DiFiore said, adding that it is still “a very gray area.”
Some stock-image libraries, such as Getty Images, have refused to carry AI-generated artwork due to the uncertainty around copyright and commercial use.
A spokesperson for the US Copyright Office told Insider that works generated only by artificial intelligence lacked the human authorship necessary to support a copyright claim.
They said the office would not “knowingly grant registration to a work that was claimed to have been created solely by machine with artificial intelligence.”
But it’s unclear whether a person entering search prompts into a program to create an AI artwork counts as a human-AI collaboration.
Representatives for Open AI said they think images generated by their programs can be copyrighted for commercial reasons.
A spokesperson from Open AI said: “When DALL-E is used as a tool that assists human creativity, we believe that the images are copyrightable. DALL-E users have full rights to commercialize and distribute the images they create as long as they comply with our content policy.”
They added that “copyright law has adapted to new technology in the past and will need to do the same with AI-generated content.”
Despite reservations, the technology’s potential also excites many artists.
Giles Christopher, a London-based commercial photographer specializing in food and drink, uses DALL-E and other AI-image generators to experiment with portraits and create artificial backgrounds for some of his commercial shots.
“I’ve come out with images that you wouldn’t question are photographs,” he said. “Some of the arguments I’ve had from photographers are that the images are looking too good.”
A portrait Giles Christopher created using artificial intelligence. Giles Christopher / Project G
He thinks the genie is out of the bottle when it comes to AI, and that artists should look for ways to include it into their work.
“I have friends in the industry who will storm out of the room if I even bring up using AI,” he said.
But he’s keeping an open mind. “I’m still on the fence. It’s like keeping your enemies close,” Christopher said.
Courtesy: Business Insider
