More from this creator
Other episodes by Kitty Cat.
More like this
If you liked this, try these.
Transcript
The full episode, in writing.
Picture this: an artist spends years perfecting their craft, only to see a computer generate an image in 10 seconds that racks up a million likes. That’s not a hypothetical—it’s the spark behind the internet’s loudest fight over what art even is: the Great A.I. Art Controversy.
A.I. art generators like Stable Diffusion and DALL-E can take a text prompt and spit out a digital painting, photo, or cartoon in seconds. They do this by analyzing billions of images scraped from the web and learning to mimic their styles. This process is called “training on data,” and most of these models have been trained on datasets numbering in the hundreds of millions of images—more visual information than any human could ever see in a lifetime.
The controversy exploded into the mainstream in 2022, when a Colorado State Fair art competition awarded first place in the “digital arts/digitally manipulated photography” category to a piece created with an A.I. tool called Midjourney. The win triggered international headlines because the artist, Jason Allen, admitted he used A.I. to create the image. Judges had awarded the blue ribbon to an entry crafted in part by algorithms, not a human hand.
What caused the uproar is that Allen’s winning piece, “Théâtre D’opéra Spatial,” was generated by entering a text prompt into Midjourney and then making minor digital edits. The AI model itself was trained on over 400 million images from across the internet, including thousands of artworks by living artists—often without their permission.
Many professional artists argue that their style, their unique visual signature, is being used without their consent. The mechanism behind this is that A.I. models like Stable Diffusion use a process called diffusion, which essentially “blurs” and then “deblurs” images, learning the patterns and elements that make up art styles, color palettes, and composition. Once trained, these models can reproduce visuals that closely resemble individual artists’ work, even if the artist never agreed to have their images included in the training set.
This led to a wave of lawsuits in the United States and Europe. Artists like Sarah Andersen, Kelly McKernan, and Karla Ortiz became plaintiffs in a high-profile lawsuit against Stability AI, Midjourney, and DeviantArt. They claim that by scraping their copyrighted work to train commercial A.I., the companies are violating copyright law, effectively building a product on stolen labor.
Defenders of A.I. art counter that the models don’t store or copy images directly. Instead, they learn statistical relationships between pixels, much like a human artist absorbs influences throughout their career. If an artist studies Rembrandt, they might paint in his style, but that’s not the same as copying a painting. Supporters argue that A.I. is just the next brush, the next camera, a tool in the creative arsenal.
One flashpoint in the debate erupted in the gaming world. In 2023, the Korean developer Pearl Abyss dropped a trailer for its game “Crimson Desert” featuring landscapes and assets generated by A.I. art tools. Fans noticed familiar visual motifs—sometimes uncanny echoes of existing game concept art—suggesting the A.I. was borrowing too closely from original creations. The controversy forced the company to clarify which assets were A.I.-generated and which were made by their human team.
Artists in the game industry responded on social media by sharing their own “ugly placeholder” art—quick sketches, stick figures, and basic shapes used internally to test games before the final art is added. They did this to make a point: placeholder art is intentionally rough, because creating polished assets is expensive and time-consuming. If A.I. art means companies can skip hiring human artists for early development, it could threaten thousands of jobs in a $180 billion global industry.
The scale of A.I. art’s reach is massive. One dataset used to train Stable Diffusion contained over 5.8 billion image-text pairs, according to technical documentation released by the developers. That’s equivalent to scraping nearly every image uploaded to popular art sites, commercial photo libraries, and even personal social media pages. The mechanism for this mass data collection is automated web crawlers, which index and download publicly visible images regardless of copyright status.
Not everyone in the art world is upset. Some digital artists, like Mario Klingemann, embrace generative A.I. as a new creative partner. Klingemann has used A.I. to create artworks that have sold at Sotheby’s and Christie’s, arguing that the “art” is in the curation—the selection and editing of images generated by machine, not just in the brushstrokes.
The debate has spilled into online platforms, too. DeviantArt, a website with over 75 million registered users, introduced an “opt-out” option for artists who don’t want their work included in A.I. training data. This required artists to explicitly mark their work to prevent scraping, but critics argue it puts the burden on artists rather than on A.I. companies to respect copyrights.
Some platforms have gone further. In early 2023, Getty Images banned all A.I.-generated artwork from its marketplace, citing concerns about copyright violations and legal liability. Getty, which licenses over 415 million images, stated that it could not verify the provenance of A.I. art, making it risky to sell to clients.
Meanwhile, the definition of “art” itself is up for grabs. Philosopher and critic John Dewey described art as a human expression, but the mechanisms of A.I. art creation challenge this by removing the artist’s hand from the process. Instead of a traditional workflow—sketch, plan, execute—A.I. art emerges from selecting, tweaking, and finessing prompts, making the artist more of a curator or an editor.
There’s even a split among audiences. Some see A.I. art as a democratizing force, enabling people without formal training to create compelling visuals. Others argue it floods the internet with mediocre, derivative images, making it harder for original human-made artwork to stand out. One estimate from a major image hosting site suggested that in a single week in 2023, over 34 million A.I.-generated images were uploaded—more than the entire previous year’s worth of photos combined.
A final twist: some A.I. art generators have been caught producing near-perfect replicas of copyrighted works when prompted with specific artist names or phrases. This happens because the models “overfit” on certain styles, reproducing details that are nearly identical to the originals, blurring the line between inspiration and infringement.