The year of our lord 2022 could be accurately described as the rise of AI. Instead of Skynet raining fire on our heads, we have AI image generators creating a different kind of apocalypse, especially for artists promoting their work online. So far, few have attempted to answer how creators can actually respond to systems that fetch their work from the Internet, using art to create new works without offering them any credit.
On Friday, DeviantArt released their new To imagine AI art generator. Based on the existing Stable Diffusion AI model, this new system will actively tag their images as AI and even credit the creators it used to create the image when they are published on the DeviantArt site.
The venerable DeviantArt’s foray into AI art is the most recent and perhaps most effective compromise between artists and AI art generators. Perhaps most important with the 22-year-old site’s announcement is that artists will have the ability to tell the AI not to use their art when creating a new image. When users upload an image, they can press a button that will prevent its own system from grabbing it for generative use. The company also said that pressing this button would create HTML meta tags that would signal to any other external system that the art was not up for grabs.
According to the company’s press release, anything that removes art from its platform when associated with this metatag would violate DeviantArt’s terms of service. Users browsing the site will also be able to control how much art labeled as AI-generated appears in their feeds.
Moti Levy, CEO of DeviantArt, told Gizmodo in a Zoom interview that this is a necessary step given that there are already tens of thousands of images labeled as AI-art on their platform. . There has also been a 1000% increase in the number of images uploaded to their site in the last four months alone.
“You can’t just enjoy the fruits of other people’s labor without crediting them or giving them the power to control their work,” he said. At the same time, he added that DeviantArt “doesn’t censor anything…it’s counterintuitive because you can’t put [this technology] back in the bottle.
What are the current issues with AI art?
Despite the popularity of AI image generators in less than a year, the advance of this technology has been accompanied by an equally rapid decline of existing image hosting platforms. So far the response from some sites has been mixed bag. Shutterstock recently announced that they are enable AI art on its platform by making it easy to download OpenAI DALL-E 2 images directly from their site. The site will then compensate users whose images were uploaded to train the DALL-E system. At the same time, Shutterstock said it banned anything created with any other AI art generator.
Moti said they weren’t quite there yet to start thinking about offering compensation to artists whose work is used for AI generation. Meanwhile, Getty Images banned AI art, and so far he seems to be sticking with that decision. Liat Gurwicz, director of marketing at DeviantArt, said banning all AI art would be “a temporary band-aid”.
“It puts a rig in a position where you have to be the judge, for example, if I used an AI tool to make the background of a room, but then drew the stay in hand, is it AI?” she says. “I think for us what’s most important is transparency between us, our tools, our technology and our community.”
One of the persistent and growing problems with AI image generators is that they essentially use the work of artists to create their new images, in most cases without any sort of permission from their original creators. These include physical and digital photos and artwork. AI generators have also proven capable of replicating live performers. Greg Rutkowski, a Polish artist known for his epic fantasy style, has been reproduced to such an extent that he fears his name, style and livelihood will be buried under the deluge of AI-generated art who iterates on his style with chilling precision.
There were several reports about Rutkowski, but other artists were both surprised and angry to learn their art was also used to train AI models. The CEO of a gaming company has been slammed after submitting AI art generated by popular AI art generator Midjourney to a local art contest and won. Stable Diffusion, one of the most popular AI art generators on the internet, was trained using the LAION image set, which dredged images from all over the internet, including copyrighted photos of Getty Images.
DeviantArt hopes other sites will follow
Of course, new DeviantArt efforts will not be able to remove any art that has been siphoned off for an existing art generator. Without any agreement with other companies, nothing can stop companies from continuing to drag user art from DeviantArt for their own creation. Moti said he started reaching out to other companies to see if they could create some sort of deal. At the same time, other image hosting sites may need to start creating a similar deal with their artists if they really want to start pressuring companies creating new AI art generators.
“We need to put pressure on them, and there needs to be cooperation from all players in this area,” the CEO said. “Any platform where artists can submit their work on the open web will have to implement this solution or something else… They can’t let their creators’ art escape the platform without letting them decide how their art will be used.”
Stability AI CEO Emad Mostaque explained that their Stable Diffusion system will only become more sophisticated and prolific over time. The CEO has been a strong advocate for unobstructed and expansive AI technology, to the point that he has fought against people who advocated preventing their system from producing violent or pornographic content.
Of course, purchasing DeviantArt’s blueprints will still cost you money. All free users get five prompts for free, though those paying for the Core subscription tier at $3.95 per month will get 50 free monthly prompts and three images per prompt. Meanwhile, those who pay $9.95 or $14.95 for the Pro or Pro+ tiers will receive 200 or 300 prompts and four images per prompt, respectively. The cost of additional prompts above the allotted number is higher for lower cost tiers than for more expensive ones.