Social media feeds have been recently flooded with AI-generated art that transforms photos of individuals, their friends, families, and pets into characters that look like they’ve stepped straight out of a Studio Ghibli film, appeared in an episode of South Park, or joined the cast of The Simpsons. With just a few prompts to ChatGPT or other generative AI tools, users can create images that mimic the distinctive visual styles of beloved animation studios.

But beneath the fun lies a complex web of legal questions. As these AI tools become increasingly sophisticated at mimicking distinctive artistic styles, important questions arise about copyright infringement, trademark protection, and other legal considerations. What potential liability do companies like OpenAI (the maker/owner of ChatGPT) have if the artists behind these styles aren’t happy with this newest trend? Do end users have any liability themselves?

Copyright Considerations: Can You "Own" an Artistic Style?

The copyright questions surrounding AI-generated art involve two main components: whether the AI tools violated copyright during training, and whether the output itself constitutes infringement.

The Idea-Expression Dichotomy

A fundamental principle in copyright law is the "idea-expression dichotomy," which establishes that while specific expressions are protected, underlying ideas and concepts are not. Artistic style generally falls on the “idea” side of this distinction.

As courts have established, a “defendant cannot be held liable for copying only ‘ideas’ or ‘concepts’ from a plaintiff's work. Instead, to be liable for copyright infringement, a defendant ‘must … copy enough of the plaintiff's expression of those ideas or concepts to render the two works 'substantially similar.’” This principle was recently reaffirmed in cases like Hanagami v. Epic Games (9th Cir. 2023) and Tangle, Inc. v. Aritzia, Inc. (9th Cir. 2025).

This means that while Studio Ghibli might potentially prove copyright infringement during an AI model’s training (if their copyrighted works were used without permission), they would face significant challenges proving that an image of a user rendered “in the style of” Studio Ghibli violates their copyright. Practically, that means while the companies who trained and developed these tools may have some exposure, end users likely do not.

Training vs. Output: Recent Cases

There are now 41 lawsuits against the owners of GenAI tools. Most of these pertain to books or other written works while only a handful concern visual works such as photographs or artwork. While decisions from those written-work lawsuits can help guide analysis in lawsuits involving other types of copyrighted material, it is not always a 1:1 application. That makes the handful of decisions from cases involving visual arts even more important. One such case is Andersen et al v. Stability AI Ltd., which is currently pending in the Northern District of California and has generated two opinions considering whether the plaintiffs’ claims state a claim under the Rule 12(b)(6) standard.

In October 2024, the court allowed the plaintiffs’ claims of direct copyright infringement to continue when the claims were based on the defendants’ copying the plaintiffs’ images to train the defendants’ LLM but not where the plaintiffs’ claims were based on the output. To show copyright infringement of two resulting products, the plaintiff must show the two products are “substantially similar.” Where a resulting image is simply in the “style of” another artist that is very difficult.

The October 2023 decision allowed the plaintiffs to amend their complaint, which they did. The defendants then moved to dismiss the amended complaint, and the same court issued another opinion in August 2024. The issue of direct infringement based on training the LLM was again at issue. And the court again held the claim could proceed. The plaintiffs included two exhibits to their complaint showing the GenAI tool produced images in the style of the plaintiffs’ art. The plaintiffs did not allege the output was infringing, however, but that this was evidence the LLM was trained on the plaintiffs’ copyrighted material. In support of its motion to dismiss, the defendant argued the output could just as easily be because the LLM was trained on unregistered works or utilized only unprotected elements from the plaintiffs’ works. The court acknowledged that could be true and, if so, would defeat the copyright claim – but held that was not appropriate to resolve at the motion to dismiss stage.

Fair Use Considerations

Fair use is an affirmative defense to copyright infringement, meaning there’s still an infringement but it is legally excused in favor of public policy/benefit. The analysis includes four factors, one being “the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes.”

While many users create these stylized images for personal, non-commercial purposes, this may not provide strong protection. The AI companies themselves are typically commercial entities, and their use of copyrighted materials for training may not align with the traditional fair use purposes of criticism, comment, news reporting, teaching, scholarship, or research.

Beyond Copyright: Trademark and Other Legal Considerations

Copyright isn’t the only legal framework at play in these scenarios. Other important considerations include:

Trademark and Trade Dress

While copyright protects original creative expressions, trademark law protects marks that identify the source of goods or services. Animation studios like Studio Ghibli may have valid trademark claims if AI-generated images create confusion about the source or sponsorship of the content.

Trade dress protection—which covers the visual appearance or image of a product and is part of the overall umbrella of trademark law—may also be relevant. In the August 2024 Andersen decision, the court allowed the plaintiff's claim for trade dress violation to continue, finding while the plaintiffs had “no protection over ‘simple, cartoony drawings’ or ‘gritty fantasy paintings,’” their trade dress claims had to be considered holistically and were allowed to proceed to at least the summary judgment stage.

False Endorsement

If a generative AI company promotes that their tool can create art in the style of specific artists or studios, this could potentially lead to false endorsement claims under the Lanham Act. In Andersen, the court denied a motion to dismiss such a claim when Midjourney’s CEO promoted a list of artists whose style their AI could recreate. This

Contract Law and Terms of Service

Many content creators license their work with specific restrictions against AI training. When these terms are violated, contract law may provide remedies where copyright law falls short. Additionally, AI companies’ own terms of service may contain relevant limitations or permissions regarding the styles users can mimic.

Conclusion: Navigating Uncharted Territory

The phenomenon of AI-generated art “in the style of” famous creators sits at the intersection of rapidly evolving technology and established intellectual property law. While current legal frameworks provide some guidance, many questions remain unresolved.

For users enjoying these creative tools, understanding the potential legal implications helps ensure responsible use. For content creators, a multifaceted approach to protecting intellectual property—spanning copyright, trademark, contract law, and beyond—offers the best protection in this new frontier.

As AI continues to advance, we can expect further refinement of both the technology and the legal principles governing its use, hopefully striking a balance that fosters innovation while respecting creators’ rights.