We are not in the prediction business. But the last time we made a prediction we were right.
That was way back on January 10, 2023 when we predicted the first class-action copyright lawsuit against AI. A few days later, on January 14, 2023, Sarah Andersen and other visual artists filed that lawsuit against Stability AI, Midjourney, and other image generators.
Well, this week it feels like déjà vu.
OpenAI released, on Tuesday, an upgrade to its image generation in the most advanced version of ChatGPT. And now social media exploded with images in the distinctive anime style of Studio Ghibli, founded by Hayao Miyazaki. Apparently, Miyazaki’s no fan of AI. He reportedly once said he was “utterly disgusted” by an AI animation in 2016 and he would “never wish to incorporate this technology into my work at all.”
examples of ghibli-style images
Sam Altman even created a profile image in Ghibli style:
The White House also posted an image in Ghibli-style depicting ICE arresting someone:
But they aren’t the only ones. Others are using the Ghibli style with memes, characters from movies, and other scenes. The flood of Ghibli-style images has spurred controversy and great media coverage, here and abroad.
So, is there a copyright lawsuit about to drop?
If we owned a farm, we would bet it on another lawsuit.
Taya Christianson of OpenAI told the New York Times in an email: “We continue to prevent generations in the style of individual living artists, but we do permit broader studio styles — which people have used to generate and share some truly delightful and inspired original fan creations.”
That distinction doesn’t hold water.
Under U.S. copyright law, corporations are treated as authors (under the work made for hire doctrine) no less than individuals. While it’s true broader styles (e.g., Cubism and Surrealism) are not copyrightable to one artist, the distinctive style of one artist, such as Saul Steinberg, who illustrated covers for the New Yorker, is a part of their copyrightable expression. Dr. Seuss was another artist who had a very distinctive visual (and literary) style.
So, yes, an artist’s own distinctive style can fall within what copyright protects. Granted, litigation will determine the scope of copyright protection for an artist’s asserted distinctive style–what the elements of the style are and which are protected by copyright. At some point, a style may be or become too broad because it constitutes a general approach or genre. And, with some types of works, such as musical works, assertions of copyright over a personal style of one musician may be problematic because it may overlap too much with elements that define heavily genre-based types of works (e.g., rap, country, pop music) that all artists use. But, where the intended target of copying is not a general style, but instead one artist’s distinctive style, that may raise a question of copyright infringement.
It’s simply false to suggest that copyright does not ever protect an artist’s style. Case law already recognizes that an artist’s distinctive style is a part of expression and may fall within copyright protection.
To think about this issue in reverse, if someone creates visual characters in the style of Dr. Seuss or illustrated magazine covers in the style of Saul Steinberg, it’s possible those style-imitative recreations infringe their copyrights. Here, since all the references to create this particular style on OpenAI’s ChatGPT are to “Ghibli” by name (just type “Ghibli style” in ChatGPT), that provides evidence that this style is distinctive to Ghibli Studios, instead of being a broad genre of anime.
Plus, as the above examples show, ChatGPT’s Ghibli-style images often include depictions of characters or people, meaning elements that are typically featured in Ghibli movies or anime. This controversy isn’t merely copying a distinctive style. Instead, it’s copying elements from the universe of Ghibli’s works and depicting them in the same distinctive style in AI outputs that may be substantially similar. In other words, the possible copying goes beyond style to include character and figure delineations by Ghibli. The dead giveaway in the ChatGPT-generated Ghibli-style images is the similarity in how especially the characters or figures are composed (e.g., the selection and arrangement of expressive, doll-like faces with round eyes with big eyeballs, thin lips, often strands of hair resembling Vs, shorter torsos, presented in desaturated, flat but warm colors, to highlight a few features). (For those more familiar with Dr. Seuss characters, consider how much easier it is to identify if someone has copied Dr. Seuss’s style if the image includes characters resembling features commonly found in Dr. Seuss’s characters.) The question is whether these similarities are substantial and based on copyrightable expression.
Indeed, Judge Orrick has already recognized in Andersen v. Stability AI that outputs of image generators “in the style” of a particular artist sufficiently alleged, to survive a motion to dismiss, copyright infringement to support the respective claims for secondary liability against the AI companies. Plaintiffs “allege that Stable Diffusion used plaintiffs’ artistic works as ‘training images’ and as a result Stable Diffusion can produce output images ‘in the style’ of those images.” Judge Orrick even allowed a false endorsement claim under the Lanham Act against Midjourney based on the allegations that it used the names of many artists in training its AI model to produce visual images in the styles of the respective plaintiff artists. People might allegedly confuse those outputs in the styles of the artists as being endorsed by them.
Whether or not the plaintiffs prevail on their claims is a different story. Judge Orrick said the claims can be tested on summary judgment. At the very least, however, the ruling does show that training an AI model to produce outputs in the style of particular artists presents a possible claim of copyright infringement to survive a motion to dismiss.
Update on Mar. 28, 3:16 PM:
According to the Times of India, “users can also create Ghibli-style photos with Grok as well as Google Gemini by either typing our detailed prompts or sharing an image with Grok and asking the chatbot to turn it into a Ghibli-style image.”
If true, that would likely mean that the same type of copyright lawsuit by Ghibli against OpenAI could also be brought against Grok and Google.
Related Stories