The recent advancement in artificial intelligence technology has been heralded as everything from a remarkable breakthrough, to a lucrative enterprise, to a seriously annoying nuisance. Artificial intelligence is currently used in many contexts and sectors, including healthcare, banking, e-commerce, manufacturing, and, more recently, the arts.
As AI is being used to generate works such as screenplays, short stories, poems, and art, it is drawing the ire of human creators and ethicists alike. Putting aside for now the more philosophical question revolving around whether machines can truly “create” art, many artists and licensors claim that the AI-generated art amounts to nothing more than copyright infringement.
One such company currently facing litigation involving this very issue is Stability AI, a startup founded in 2020 by Emad Mostaque. In August 2022, Stability AI launched Stable Diffusion, an AI model that generates nearly instantaneous images and artwork based on prompts keyed in by the user.
Stable Diffusion’s website describes itself as “a latent text-to-image diffusion model capable of generating photo-realistic images given any text input.” The website hosts a free “playground” where anyone can experiment with generating virtually any type of image with just a few keywords. So for example, when one types in “pug astronauts” as a prompt, the AI model may yield results similar to the following:
Images generated with Stable Diffusion AI
However, this ability to immediately generate bespoke images does not simply appear out of a vacuum. An AI system must have access to vast amounts of data in order to learn and understand patterns, thereby refining its own performance and accuracy in the process. In the case of Stability Diffusion, this entails scraping the internet for enormous quantities of images and their accompanying information in order to train the model to understand how images and associated text are related.
Getty Images, a preeminent provider of online digital media, has taken issue with Stable AI’s use of copyrighted images to train the Stable Diffusion model. On February 3, 2023, Getty Images filed a complaint against Stability AI in the District of Delaware that claimed the AI company unlawfully used over 12 million of Getty Images’ photos and accompanying captions and metadata.
Getty Images hosts an enormous library of digital content, the majority of which are copyrighted stock photographs that customers (primarily creative, corporate, and media professionals) can pay to use, royalty-free. Getty Images obtains its high-quality images by employing staff photographers and licensing the use of images from third parties.
The complaint asserts that Stability AI’s use of Getty’s database to train the Stable Diffusion model so that it can synthesize new images violates Getty’s intellectual property rights. The suit alleges that Stability AI has taken “[proprietary] content from Getty Images without permission, depriving Getty Images and its contributors of fair compensation.”
The complaint points out that since Stable Diffusion was trained using Getty Images, many of the visual works that Stable Diffusion generates are derivative of the original copyrighted images owned by Getty. The ability to create derivative works is an exclusive right endowed to the copyright owner. The suit further states that “…by utilizing Getty Images’ copyrighted content for artificial intelligence and machine learning, Stability AI is stealing a service that Getty Images already provides to paying customers in the marketplace for that very purpose.”
These claims are not the first that Stability AI has faced in 2023. On January 13th, three individual artists (Sarah Andersen, Kelly McKernan, and Karla Ortiz) filed a separate class action suit against Stability AI, along with defendants Midjourney, Inc. and DeviantArt, Inc.—two other AI platforms that utilize Stable Diffusion’s model.
The suit was filed on behalf of all artists “that own a copyright interest in any work that was used to train any version of an AI Image Product that was offered directly and/or incorporated into another product by one or more Defendants.” The class action refers to Stability Diffusion as a “21st-century collage tool” where “the resulting image is necessarily a derivative work, because it is generated exclusively from a combination of the conditioning data and the latent images, all of which are copies of copyrighted images.”
While Stability AI has not formally responded to the lawsuits as of the time of this writing, several news outlets have reported a company spokesperson as stating: “Anyone that believes that this isn’t fair use does not understand the technology and misunderstands the law.”
Fair use is a fuzzy concept in copyright law, without a solid definition. The doctrine allows for the use of unlicensed, copyrighted content in certain situations. The Copyright Act sets out a series of factors for determining whether fair use has occurred. One of the factors courts consider is how much the new work replaces the market for the old work. Getty is making the case that the new AI works replace the market for the original Getty works; in fact, Getty argues that Stability Diffusion’s use of its images to train AI to create a new library of art threatens Getty’s entire business model.
Over the years, courts have adopted an additional test for determining whether the new work is “transformative”—whether it adds something new, with a further purpose or different character, altering the first with new expression, meaning, or message. The doctrine of transformative use has expanded in recent years, perhaps most notably through the cases challenging the Google Books project, wherein the court found that using unpublished books to create an indexed search tool was a transformative, and therefore fair, use. The current AI cases wherein existing art is mashed together to create new art may well teach us more about how far the transformative use doctrine can go.
If you have any further questions regarding intellectual property rights and copyright issues, please contact Tara Aaron-Stelluto.
