When Jason Allen won first prize in the Colorado State Fair’s art competition in 2022 for his “Théâtre D’opéra Spatial” piece under the “digital arts/digitally-manipulated photography” category, he ultimately challenged the art community on what it means to be an artist. Allen stated that “he made clear that his artwork was generated with Midjourney” throughout the art competition’s duration—including the “drop-off” and “his narrative submission.” The Colorado State Fair defines “digital art” as the “artistic practice that uses digital technology as part of the creative or presentation process”. Even though the art competition’s judges, Cal Duran and Dagny McKinley, were not aware of the fact that Allen used AI to create his winning art, they stated that it “wouldn’t change their judgment.” The fact that Allen’s AI-generated art won turned the art world upside down, propelling him to go viral on social media and garnering national attention. However, Allen’s win also brought on the concerning question of whether images generated by AI can be “art.”
Throughout Allen’s process of creating “Théâtre D’opéra Spatial,” he did not use any digital painting tools to create his work—instead, he used words. To craft his prize-winning work, Allen would input a variety of prompts into Midjourney AI to generate numerous renditions of the image. According to Allen, it is more difficult than one may expect as he had to use a “limited number of words” that included “key phrases that affected the lighting, perspective, composition, atmosphere, subject, and other attributes.” After going through various reiterations of the prompt over 900 times, Allen chose his top three renditions for submission.
With the rise of AI-generated art through programs such as DALL-E2 and Midjourney becoming accessible to the public for use, the reception to Allen’s win was imbued with anger from the art community. One Twitter user commented: “We’re watching the death of artistry unfold before our eyes… If creative jobs aren’t safe from machines, then even high-skilled jobs are in danger of becoming obsolete. What will we have then?” However, the use of AI-generated art is becoming prevalent, ranging from artists like Allen to journalists and large corporations. An example of this is when Atlantic writer Charlie Warzel used a Midjourney-generated picture of Alex Jones in the magazine’s newsletter. In response to Warzel’s usage of AI-generated art, cartoonist Matt Borrs said: “Technology is increasingly deployed to make gig jobs and to make billionaires richer, and so much of it doesn’t seem to benefit the public good enough… AI art is part of that. To developers and technically minded people, it’s this cool thing, but to illustrators it’s very upsetting because it feels like you’ve eliminated the need to hire the illustrator.”
And so, this elimination of “the need to hire the illustrator” essentially dismantles the connection between the art and the artist as now the artist is seen as merely a middleman.
How AI image generators work
After launching their respective AI image generators to the public in 2022, both Midjourney and Stability.AI are enjoying more than 10 million users. However, the existence of these AI image generators represents the graveyard of jobs and theft for digital artists; these AI image generators are trained by “scraping” billions of images from the Internet—including works by digital artists who did not consent their works to be scrapped. Dazed Magazine’s Thom Waite defines scraping as “creating software that automatically collects data from various sources, including social media, stock image sites, and (maybe most controversially) sites where human artists showcase their work, such as DeviantArt.”
One common dataset used by these AI image generators is a dataset made by German charity LAION that “provides datasets, tools, and models to liberate machine learning research,” where the software looks for “image-text pairs” before having them be compiled in a multitude of datasets. LAION also explicitly states that their datasets are “simply indexes to the internet, i.e. lists of URLs to the original images together with the ALT texts found linked to those images.” From this description of LAION’s statement on their datasets, they are merely taking what is posted on the Internet—ranging from images, videos, text files, etc.—and compiling it into a massive collection of datasets. To “train” these AI image generators, the text-image pairs from these datasets create their own “knowledge base to teach image generators how to ‘create’ images for them.” By doing so, this allows for the AI to “make connections between the composition and visual data of an image, and the accompanying text.”
Beyond these newly made connections, the AI undergoes “diffusion,” a process where the AI is “shown increasingly of blurry or ‘noisy’ images, and taught to reconstruct the original image out of the visual noise.” After several times, the AI possesses the ability to create images that are unlike any of the images it has trained on—which is only made possible through its process of copying billions of images from the DAION dataset.
The problem of consent in generative AI
As digital artists share their works on the Internet for free—social media platforms like Twitter and Instagram, websites decided to share art—to gain traction and attention, it is common for them to be concerned with plagiarism. However, their fears are elevated with the rise of generative AI. Stable Diffusion states explicitly in its FAQ (“Frequently Asked Questions”): “There was no opt-in or opt-out for the LAION 5b model data. It is intended to be a general representation of the language-image connection of the Internet.” Even if the LAION 5b model was meant to be a “general representation of the language-image connection of the Internet,” many artists have spoken out in concern about the conspicuous theft of their hard labor like Canadian illustrator and content creator Sam Yang.
Yang, who has a total of more than 3 million followers across YouTube, Instagram, and Twitter, is among those whose works have been replicated by various AI image generators to the point in which he had received a malicious email from a stranger to determine “which custom AI image generator best mimicked his own style.” To replicate Yang’s distinctive style of characters with “Disney-wide eyes, strawberry mouths, sharp anime-esque chins,” all people needed was around 100 pictures of Yang’s work and upload them into their choice of AI image generators for training. In an attempt to understand why people were replicating his style without consent, Yang resorted to the very place that launched his fame: the Internet. He soon found out that it was because AI image generators gave everyone and anyone the opportunity of not needing to contact the artist for permission to replicate their style. Some internet users’ justification for using these AI image generators was that if AI companies such as Midjourney and Stable Diffusion had already scraped the Internet to train these AI generators, why could they not do the same?
However, this lies in the reality of working as a digital artist rather than a traditional artist. These digital artists do not make their living off the traditional sense of an artist where they would showcase their works in exclusive art galleries or auction houses, but they make their living off promoting their works on the Internet via tutorials and selling merchandise. Unlike the traditional conception of an artist where there will be patrons who buy the artist’s works, many digital creators will send merchandise that features his works such as t-shirts, posters, and keychains.
Emerging lawsuits against AI companies by creators
Because illustration rates have stagnated and decreased since the 1980s, many artists resort to selling the usage rights of their artworks to make a living. Deb JJ Lee, a Brooklyn-based illustrator, further attests to the decreasing wages for those involved in the art community: “I know freelancers who are at the top of their game that are broke, I’m talking [illustrators who do] New Yorker covers. And now this [referring to the rise of AI image generators]?” After finding out from fellow artists that a lot of these AI image generators were trained on the LAION dataset, it became increasingly apparent that “[a]lmost every digital artist has images in LAOION, given that DeviantArt and ArtStation were lifted wholesale, along with Getty Images and Pinterest.”
Such a discovery led rise to a lawsuit against Stability.AI and Midjourney by a trio of artists—Sarah Andersen, Kelly McKernan, and Karla Ortiz— “allege that these organizations have infringed the rights of ‘millions of artists’ by training their AI tools on five billion images scraped from the web ‘without the consent of the original artists.’” Artists are not the only ones filing lawsuits against these AI companies as Getty Images, the famed media company, also filed a similar lawsuit against Stable Diffusion. Craig Peters, Getty Images’ CEO, stated: “The company [Stability.AI] made no outreach to Getty Images to utilize our or our contributors’ material so we’re taking an action to protect our and our contributors’ intellectual property rights.” However, in the lens of these AI companies, the act of “scraping” is “covered by laws by the U.S. fair use doctrine,” but those whose works have been scrapped allege that “scraping” is violating their copyright.
Waxy’s independent analysis of the images used to train one particular AI image generator—Stable Diffusion—found that Getty Images and other stock images sites are a major part of Stable Diffusion’s content to the point in which Getty Images’ watermark is often replicated in its AI-generated images. Due to the sheer controversies surrounding AI companies’ scraping without consent, they have started to make concessions to digital creators. An example of this is Stability.AI’s statement that digital artists and creators will be able to “opt out of the next version of Stable Diffusion.”
Although opting out is one solution to ensure that AI image generators are not trained on copyrighted and unconsented creations, Harvard Business Review remarks that this decision deliberately places the work “on[to] content creators to actively protect their IP, rather than requiring the AI developers to secure the IP to the work prior to using it.” Considering that these artists already have a lot on their plate such as promoting their work online and ensuring that their creations are not plagiarized, this constant alert for new AI generators to opt-out of is not sustainable as there will be a continuous influx of various AI image generators. And so, these AI companies must have the initiative to require the creator’s opt-in instead of having them opt out and be mindful of incoming AI generators that may ask for their decision.
The legal future for generative AI companies
It may seem as though these lawsuits against these AI companies are obvious infringements of copyright since AI image generators were essentially trained on these billions of images without their creators’ consent; however, the lawsuits filed by both the trio of artists and Getty Images are focused on attempting to “apply existing legal standards made to protect and restrict human creators, not a borderline science-fiction computing tool.” Getty Images’ CEO Peters explicitly dismisses claims about “damages” and “stopping the distribution of [AI image generators],” but aims to “[build] generative models that respect intellectual property.” Peters continues to speak on his rationale behind his company’s lawsuit to the Verge:
“I equate [this to] Napster and Spotify. Spotify negotiated with intellectual property rights holders—labels and artists—to create a service. You can debate over whether they’re fairly compensated in that or not, but it’s a negotiation based on the rights of individuals and entities. And that’s what we’re looking for, rather than a singular entity benefiting from the backs of others. That’s the long-term goal of this action.”
As the “diffusion” process of the AI training process is a crucial aspect as it fundamentally copies and recreates images from the datasets, the artists’ lawyers argue that these AI image generators essentially “call back to the dataset” and compile millions of these images to create the image that is requested by the user, sometimes “with the explicit instruction to recall the style of a particular artist.” Consequently, the creation made by the AI image generators is deemed as a “derivative work,” a work that is not “significantly transformed” from its original material. Hence, the lawsuit filed by the three artists and their lawyers focuses on “protect[ing] human artists”; considering that once the AI image generator’s prompt has the artist’s name, the resulting AI can be seen as “derivative” even if the work is a culmination of the millions of images in the dataset. Essentially, Art in America pinpoints that this particular lawsuit by the three artists is attempting to “establish copyright over style, something that has never before been legally protected.”
However, the U.S. Copyright Office (USCO) released its policy on AI-generated works and ultimately ruled that AI-generated images from text cannot be copyrighted. “They identify what the prompter wishes to have depicted, but the machine determines how the instructions are implemented in its output,” says USCO in its guide sent to the Federal Register. According to Engadget, the USCO determines whether the AI-generated image can be copyrighted based on “whether the model’s contributions to the work are the result of ‘mechanical reproduction’ (i.e., generated in response to text prompts) or if they represent the author’s ‘own mental conception.’” This is not to say that all AI-generated images will not be copyrighted, but rather it will be on a case-by-case basis determined by the USCO through an understanding of “how the AI tool operates and how it was used to create the final work.” As the USCO’s current ruling protects artists’ usage rights in an unsettling time for their industry, this decision also puts them at a disadvantage concerning their demand for financial compensation at the hands of these AI companies.
Since generative AI continues to grow in the future, artists are not the only ones paying close attention to the shifts in regulations—but also lawyers globally. Due to the current nature of generative AI, regulations are being considered to set forth a common and global standard held by all countries, but namely China, the U.S., and Europe as the primary examples. As these platforms are trained on the Internet and a multitude of datasets, many intellectual property (IP) lawyers are grappling with the potential of unknowingly infringing on some of these datasets’ copyright. At the same time, IP lawyers are also attempting to identify the best steps in this ambiguous environment while staying alert to any AI-related regulations set forth concerning copyright law.
“If I need to comply with both U.S and EU law, and one of them lets me do it, and the other one doesn’t, I still can’t do it,” remarks Mark Lemley, a Stanford Law School professor and counsel at Lex Lumina, the firm defending Stability.AI and other generative AI companies. With regards to the EU’s track record of internet regulations like the Digital Markets Act—an act aiming to undermine the market power of Big Tech, it is alleged that they are drafting regulations on an “AI act that is due to be finalized next year.” Meanwhile, the U.S.’ “fair use doctrine” with generative AI will be carefully re-examined with the lawsuits by Getty Images and the three artists, potentially setting a precedent on how copyright law will proceed with the existence of AI image generators.
Generative AI’s disruption has catapulted the world into disarray concerning copyright law and what qualifies as original, prompting many to question which rules will be effective and generative AI’s legality. Hence, it is vital that those who use and/or develop generative AI are mindful and cautious about what data is used to generate their end-products, whether the information such as photos, texts, and videos in the used dataset is not copyrighted, and most importantly, if there are “guardrails [held] already in place to prevent 3rd party copyright infringement claims.”
If you have any questions or concerns about this insight, please contact us at firstname.lastname@example.org