OpenAI's popular AI chatbot, ChatGPT, recently became the center of attention due to a sudden user trend: a massive influx of users generating images in the style of Studio Ghibli, the legendary Japanese animation studio. This imitation trend spread rapidly, with OpenAI CEO Sam Altman even jokingly referring to himself as a "Ghibli-style twink" on social media.
However, this seemingly harmless trend took a sharp turn. OpenAI began to curb the phenomenon, implementing confusing adjustments that now frequently reject user requests to generate Ghibli-style images.
This viral trend and OpenAI's chaotic response have reignited intense debate surrounding copyright and the use of generative AI. Critics argue that this directly undermines the work of human artists and publishers, including Ghibli's talented animators.
According to Rob Rosenberg, a former Showtime general counsel and AI expert, Studio Ghibli might actually have grounds to take legal action against OpenAI. While he notes that whether this constitutes sufficient legal grounds is a "complex issue," Ghibli "might be able to claim that OpenAI violated the Lanham Act," which provides legal recourse for claims related to false advertising, trademark infringement, and unfair competition.
Rosenberg argues that Ghibli could contend that by allowing users to transform photos into a "Ghibli style," OpenAI is harming Ghibli's trademark goodwill, using Ghibli's recognizable unique style without authorization, and potentially causing consumer confusion, leading people to believe the feature is endorsed or authorized by Studio Ghibli. Furthermore, he adds that OpenAI could also face litigation for "training its large language model on copyrighted works of Studio Ghibli."
In fact, OpenAI already faces multiple lawsuits over its method of scraping data from the internet, including vast amounts of copyrighted material, for model training. Just this week, a federal judge rejected OpenAI's request to dismiss a copyright lawsuit filed by the New York Times. The newspaper, in late 2023, accused OpenAI of using its news material without permission to train its AI model, infringing on its copyrights. Mark Zuckerberg's Meta faces similar accusations of illegally downloading pirated books to train its AI models.
However, there's no clear legal precedent yet to determine whether OpenAI has violated US copyright law. OpenAI maintains that "training their models is fair use under copyright law," and therefore not infringing. This "fair use principle" has been a central point of contention surrounding whether AI companies like OpenAI are infringing on copyrights.
It remains unclear whether Studio Ghibli will pursue legal action. The company told Japanese news outlet NHK on Friday that a legal document circulating online claiming to be a cease-and-desist notice from the studio is fake. But Rosenberg cautions that it's unclear whether such lawsuits would be successful, as most courts haven't yet ruled on the validity of similar copyright claims. He believes that if OpenAI markets its platform as a place where consumers can transform photos into "Ghibli-style" animation, this would more strongly support a court ruling that OpenAI's actions violate the Lanham Act.
Rosenberg predicts that unless courts rule "consistently" in favor of OpenAI, finding that using copyrighted material to train AI models is indeed fair use, the future might see a system where AI developers need to compensate copyright holders and give them attribution when their content is used to produce output.
Beyond potential copyright issues, OpenAI's feature could also deprive Studio Ghibli of other monetization opportunities. Rosenberg points out that if Studio Ghibli plans to launch its own tool in the future allowing fans to transform photos into its signature style, OpenAI's feature has effectively preempted that market. Worse, the technology could even lead to people creating entire animation projects in the Ghibli style and falsely promoting them as genuine Studio Ghibli works, blurring the line between homage and outright misrepresentation.
