When AI generative art exploded into our lives last summer, what initially seemed like a promising creative tool quickly proved problematic. Artists were shocked to find their names being used as hints to churn out pastiche works that closely mimic their signature styles, sometimes even carrying garbled versions of their own signatures.
Even when AI-generated images don’t look much like an artist’s work, many also find it infuriating that their images have been used as training data without their consent to create a product that could threaten their livelihoods ( see our selection of the best AI art tutorials and our guide on how to use DALL-E 2 for more on text-in-image generators).
Outrage over artists’ work being scraped to train AI art generators like Stable Diffusion and Midjourney caused many digital artists to protest on ArtStation when that platform began hosting AI-generated art. While we still look in amusement at the strangest AI images, this technology is not becoming a joke for many digital artists. Below I talk to a number of artists about the issues with AI art and please note that none of the artwork in this article is AI generated.
AI Generative Art: The Problem
Illustrator Deb JJ Lee describes the moment they first encountered AI art in their style. “The experience of seeing someone pitch my work in Stable Diffusion and get a model out of it has been nothing short of devastating,” she says. “An artist’s voice is a lifetime development that is an amalgam of their visual vocabulary, interpretations of influences, and even their beliefs. That’s it. Bringing the work out of a mindless machine learning model that looks like mine, but without the struggle of developing that voice, it’s a slap in the face.”
JJ Lee isn’t the only one experiencing this, and some rights holders are reacting. Three artists filed a class action lawsuit against Stable Diffusion with a lawyer also suing GitHub CoPilot, an AI-powered programming tool trained on publicly available code from GitHub.
The lawsuit claims that “Stable Diffusion contains unauthorized copies of millions – and possibly billions – of copy-protected images” and seeks to use existing copyright law to sue Midjourney, Stability AI and DeviantArt. Getty Images also filed a lawsuit against Stable Diffusion, and the Concept Art Association raised more than $200,000 to pay a US government lobbyist, with plans to update laws to ensure artists’ interests are protected.
For individual artists, taking a legal first step can be constructive, if only to combat frustration and promote mental well-being. Roena Rosenberger, an illustrator and concept artist based in Germany, used her legal expenses insurance to hire a lawyer to sue companies that used images of her to train AI models without consent from she.
“It absolutely gave me peace of mind,” she says. “I’m someone who needs to have a plan and it helped me a lot to contact a lawyer.” Strategically it makes sense for her to wait until the German courts are more familiar with the matter before moving forward, so at the moment she’s biding her time.
Rosenberger describes the technology as “a nice shiny toy” that’s alarmingly good at mimicking artists’ styles, and notes that similar tools are likely to start influencing other professions. When that happens, many more people will join the fight to protect creators’ rights. “Our boat will not stay as small as it is now,” she says.
The art of generative AI must be ethical
Many artists hope that creators of AI image generators will be forced to discard models trained on datasets that contain work they don’t have permission to use and build new ones using images for which they have permission. Some companies are already going down this path, for example the photo and video editor Prequel (opens in a new tab) it feeds some of its effects with its own internal neural network built on images that artists have been commissioned and paid to create.
Hopefully, when artists’ rights are respected, the conversation can shift to exploring how this technology can be used to assist the creative process, rather than replace it.
“I would like to see studios having a conversation with their artists about how AI can be implemented in a way that takes the load off repetitive, boring and unwanted tasks so artists have more time for the actual important creative work work,” says Christina Kraus, a freelance artist specializing in fictional characters.
Alongside the ethical question of artists’ work being used without permission is the question of the extent to which AI art poses a threat to artists’ livelihoods. It’s often pointed out that AI can’t do the kind of problem solving that’s critical to the role of a professional artist, so do they really need to care?
“I think they have the potential to become a serious threat to working artists in the future if they aren’t regulated,” says Kraus. “We already see AI images being used on book and magazine covers, to illustrate articles, and even Netflix experimenting with the use of AI for its Dog and Boy anime.”
Kraus expects more companies to follow in Netflix’s footsteps, especially as the technology becomes more capable and overcomes some of its limitations. “It’s always about saving money and churning out as much content as possible, often with no regard for the quality of the output or the well-being of the people creating it,” he says.
“I don’t think AI can completely displace artists, but the climate will change and many of the grassroots jobs that provide early work experience will be taken over by AI. Students and artists looking to enter will have a moment yet harder”.
Kraus supports efforts to make training sets ethical by gaining consent and paying performers, but still maintains reservations about the basic nature of the technology. Could we end up with a situation similar to music streaming, where artists get paid a pittance for “using” their work?
“If AI becomes ethical, there’s nothing stopping companies from using it. If we’re lucky, we’ll get some money for our pictures in the training data and that’s it. That’s why it’s so important for studios to have a dialogue with artists instead of just cutting costs and getting rid of them. In the end, I’m not sure if ethical AI would make it better or worse for artists in general.”
The impact of AI art is severe
One artist who has become a prominent voice in the conversation is Greg Rutkowski, after it became a trend to use his name as a hint to generate a particular style of painting. “I was shocked to find so many artworks signed with my name that weren’t mine,” he says.
Rutkowski sees the use of artwork to train AI models without permission as a “many-level infringement” and believes the creation of datasets for AI should be strictly regulated. Like many artists, he notes that while AIs might be good at copying a style and producing work with superficial appeal, they can’t design art like a human does and adapt to feedback.
“There’s a purpose to designing something from scratch,” she says. “When you design clothing for a specific time period, you think about its purpose and historical context. When you create concept art of machinery, for example, it needs to make sense, be believable, and look like it works. So far, AIs produce abstract ideas, almost surreal, that may take you somewhere but have no coherent purpose.”
The current situation is something we have seen many times before and will no doubt see again: a new technology emerges and the law has to catch up. Only when that happens will we have an idea of how AI art will fit into our lives.
This article originally appeared in Imagine FX #226. If you enjoyed this piece, you will love ImagineFX. The world’s favorite digital art magazine is on sale in the UK, Europe, USA, Canada, Australia and more. A limited number of ImagineFX print editions are available for delivery to over 120 countries from our Online shop (opens in a new tab) (shipping costs are included in all prices).
#stop #generative #art