The Battle Lines
Since image generation tools entered mainstream awareness in 2022, the conflict between working artists and AI companies has only intensified. Through 2023 and into 2024, what began as online skirmishes over the ethics of training data evolved into full-scale legal battles, legislative efforts, and a fundamental reckoning with what we mean when we talk about creative labor.
The Artists' Position
The core grievance is straightforward and legitimate: AI image generators were trained on billions of images scraped from the internet without the consent or compensation of the artists who created them. Illustrators, concept artists, photographers, and painters found their distinctive styles replicated by systems that had ingested their work. The result was a technology that could produce approximations of their output in seconds, available to anyone for pennies or nothing.
Class action lawsuits were filed. On January 12, 2023, artist Sarah Andersen and several colleagues filed suit in federal court against Stability AI, Midjourney, and DeviantArt, focusing specifically on the LAION dataset, a collection of 5 billion images scraped from the internet and used to train Stable Diffusion and related systems. Artists organized collective action through groups that grew from small Discord channels into significant advocacy organizations. The emotional dimension of the fight was as important as the legal one. For many artists, seeing their visual language reproduced by a machine that had never held a brush or spent years developing technique felt like a violation that went beyond copyright.
The Getty Images Case
Getty Images launched separate proceedings in the High Court of Justice in London against Stability AI, alleging that the company copied more than 12 million photographs from Getty's collection, along with associated captions and metadata, without permission or compensation. The evidence was striking: Stable Diffusion's output sometimes included distorted versions of Getty's watermark, making the source of the training data impossible to credibly deny. That detail, a ghost of a watermark appearing in AI-generated images, became one of the most concrete and damaging pieces of evidence in the early legal proceedings.
The Technology Companies' Response
The companies building these tools offered arguments ranging from the technical to the philosophical. They claimed that AI learning from images was analogous to human artists learning from existing work, a comparison that most working artists rejected as fundamentally dishonest. Human artists absorb influence over years of active looking and practicing. An AI model ingests billions of images in hours without developing the understanding of context, intent, or craft that transforms influence into original work. Some companies began offering opt-out mechanisms, though critics pointed out that asking artists to individually remove themselves from datasets that had already been built was an inadequate solution. The data was already inside the system. The opt-out was symbolic.
The Legal Landscape in 2024
By early 2024, courts in multiple jurisdictions were working through cases that will establish precedent for decades. The fundamental legal questions being adjudicated are genuinely novel: whether training an AI model on copyrighted images constitutes copyright infringement, whether the output of those models infringes on the styles of artists in ways that existing copyright law recognizes, and whether the technology companies' fair use arguments hold when the scale and commercial purpose of the training are taken fully into account.
Stability AI's core defense rested on the argument that training is transformative use, that the model does not reproduce images but learns patterns from them. The counter-argument is that transformative use has historically required some creative act of transformation on the part of the party claiming it. Feeding someone else's work into a system designed to commercially profit from it, while contributing no creative labor of your own, is a novel interpretation of transformation.
What Is Actually at Stake
This is not a Luddite reaction to new technology. It is a demand that creative labor be respected and compensated. The artists leading this fight are not opposed to innovation. They are opposed to an innovation model that treats their life's work as free raw material.
The livelihoods at stake are real and quantifiable. Concept artists, illustrators, and visual development professionals in film, gaming, and advertising reported significant drops in commission volume through 2023 as companies substituted AI generation for work they had previously paid humans to do. The technology's commercial deployment happened before the legal and ethical framework caught up, which is precisely the pattern that the technology industry has repeated across multiple cycles of innovation. Move fast, establish facts on the ground, negotiate terms from a position of market dominance later.
That pattern deserves to be named plainly. The artists naming it are right.
The 2024 Legal Landscape
The class action lawsuits filed in 2023, including Andersen v. Stability AI and related cases against Midjourney and DeviantArt, moved through the courts in 2024 with mixed results. Some claims were dismissed on narrow technical grounds, others survived to discovery. The central question, whether training on copyrighted images without license constitutes infringement, has not been definitively resolved. The courts have moved slowly, the technology has moved fast, and the gap between the two has allowed the industry to establish commercial dominance before the legal framework caught up.
The Getty Images case proceeded separately, with the watermark evidence providing the clearest visual demonstration of what training on scraped data actually produces. When an AI image generator reproduces the visual artifacts of a specific stock photography service's watermarking system, the argument that training constitutes a transformative use becomes substantially harder to sustain.
The opt-out mechanisms that AI companies offered as a concession to objecting artists were widely criticized as inadequate: they placed the burden on individual creators to register their work, required using specific tools designed by the same companies profiting from the training, and applied only prospectively to future crawls rather than retroactively to data already incorporated. An opt-out requirement that arrives after the data has already been taken and the commercial product already built is not an ethical remedy. It is a public relations gesture.
