DeviantArt, the artist community owned by Wix, today announced a new protection for creators to prevent art-generating AI systems from being developed using their artworks. An option on the site allows artists to prevent third parties from scraping their content for AI development purposes, with the aim of preventing work from being swiped without the artist’s knowledge or consent.
“AI technology for creation is a powerful force that we cannot ignore. . . . It would be impossible for DeviantArt to try to block or censor this art technology,” CEO Moti Levy told londonbusinessblog.com in an email interview. “We see so many cases where AI tools help artists’ creativity, allowing them to express themselves in ways they couldn’t in the past. That said, we believe we have a responsibility to all creators. To support AI art, we also need to implement fair tools and add safeguards in this area.”
When AI-generated artwork began to spread across the web earlier this year, fueled by the release of text-to-image tools such as Stable Diffusion and DALL-E 2, art-housing platforms were forced to take a policy stance. Some, including Newgrounds, PurplePort, and Getty Images, banned AI-generated art altogether, regarding both the impact on artists and the legal consequences of art created with tools developed on copyrighted works.
Today’s advanced AI art tools ‘learn’ to generate new images from text prompts by ‘training’ on billions of existing images, often drawn from datasets scraped together by public image hosting websites such as Flickr and ArtStation. Some legal experts suggest that training AI models by scraping public images — even copyrighted ones — will likely be covered by: fair use doctrine in the US But it’s a matter that is unlikely be settled soon – especially in light of conflicting laws being proposed abroad.
OpenAI, the company behind DALL-E 2, took the proactive step of licensing some of the images in DALL-E 2’s training dataset. But the license was limited in scope and rivals have so far not followed suit.
“Many creators are rightly critical of AI generation models and tools. First, they don’t give creators control over how their art can be used to train models, nor do they let creators decide whether to allow their style to be used as a source of inspiration in image generation,” Levy continued.” As a result, many creators have seen AI models train with their art or worse: AI art is generated in their style with no option to opt out or get proper credit.”
Art created with DeviantArt’s DreamUp tool. Image Credits: Digitonaut / DeviantArt
DeviantArt’s new protection is based on an HTML tag to prevent the software robots that crawl pages for images from downloading those images for training sets. Artists who indicate that their content cannot be used for the development of AI systems will have the guidelines ‘noai’ and ‘noimageai’ added to the HTML page associated with their art. To remain compliant with the updated DeviantArt terms of service, third parties using DeviantArt content for AI training must ensure that their datasets exclude content that contains the tags, Levy said.
“DeviantArt expects all users who visit our service or the DeviantArt site to respect the creators’ choices about the acceptable use of their content, including for AI purposes,” Levy added. “When a DeviantArt user does not consent to the use by third parties of their content for AI purposes, other users of the service and third parties accessing the DeviantArt site are prohibited from using such content to create an AI system, as input to previously trained AI system, or to make a derivative copy available, unless use of that copy is subject to terms at least as restrictive as those set forth in the DeviantArt Terms of Service.”
It’s an attempt to return power to artists like Greg Rutkowski, whose classic painting styles and fantasy landscapes have become one of the most used prompts in the AI art generator Stable Diffusion – much to his chagrin. Rutkowski and others have expressed concern that AI-generated art that imitates their style will displace their original works and hurt their income when people start using AI-generated images for commercial purposes.
The tools have caused a lot of controversy in recent months. A system Trained to imitate the style of acclaimed South Korean illustrator Kim Jung Gi, who passed away suddenly in early October, was condemned by many in the art community as a tacky stunt. After to win an award at the Colorado State Fair art competition, artworks created by AI caused a fierce response. Elsewhere, character designers such as Hollie Mengert criticized what they see as poor AI imitations of their style that are nevertheless inexorably tied to their name.
On the DeviantArt side, it is encouraging creator platforms to adopt artist protections and says it is already in discussions about “multi-player” implementation. But it’s unclear whether it will be able to rally the wider industry behind its approach; less scrupulous actors could theoretically ignore DeviantArt’s terms of service to scrape images regardless of the HTML tag. Technologists Mat Dryhurst and Holly Herndon are leading a separate effort called Source+ to allow people not to allow their work or likeness to be used for AI training purposes. Meanwhile, Shutterstock is banning all AI art not created with DALL-E 2 in order to reduce copyright issues (and likely to preserve its partnership with OpenAI).
Image Credits: Digitonaut / DeviantArt
Unlike Shutterstock, DeviantArt has allowed — and will continue to allow — art generated with third-party AI tools on its platform, Levy says, though it encourages users to upload AI-generated art to view it as such. to label. He claims that tens of thousands of images tagged as “AI art” are submitted to DeviantArt every month, growing more than 1,000% over the past four months.
“Since the founding of DeviantArt, we have never believed in blocking art genres or categories. We have always made room for and support for all kinds of creators and their works,” said Levy.
In addition to simply allowing AI art, DeviantArt commits to supporting it through a new in-house AI art generator, Invent, which Levy says is designed to enable “safe and fair” AI image generation. DreamUp is built on Stable Diffusion and uses DeviantArt-specific models to guide the generation process into styles that are often trending on the platform.
When it launches this week, DreamUp will be offered as part of DeviantArt’s premium Core plans, which start at $3.95 per month. All DeviantArt members can try the tool with up to five free prompts.
Levy didn’t say whether DreamUp will automatically filter out subjectively offensive content like graphic violence and gore, similar to DALL-E 2 and most other commercial AI art tools. But he noted that art produced by DreamUp will be bound by DeviantArt’s terms of use and etiquette policy, which prohibits deepfakes, hateful images, and explicit art.
Images created by DreamUp are automatically tagged as “#AIart” on DeviantArt and contain a visible watermark. In an effort to honor the artists whose works were used to train DreamUp, DeviantArt will showcase the styles that inspired the art generated by DreamUp and link to the artist’s usernames where appropriate.
Image Credits: DeviantArt
Levy acknowledges that some users prefer not to see AI-generated art on a platform like DeviantArt, and says account holders can set preferences to hide all images tagged #AIart. “We are committed to continue researching and learning from new creator and web technologies, but most importantly, we continue to listen to all users to understand what they want and need to grow and succeed on their creative journey he added.