The Character Without an Author: Character Copyright Protection in the Age of Generative AI

This article is part of DWT's The Generative Slate series. It explores the use of generative AI in the production and distribution of content.

In recent years, courts have increasingly grappled with an important question for rightsholders and content creators: When can a character in an expressive work—such as a film or television series—qualify for copyright protection independent of the work in which that character appears? Well-known characters like Rocky, Godzilla, E.T., and James Bond have all received character copyright protection from courts around the country. Others have not. Just last year, in Carroll Shelby Licensing v. Halicki, 138 F.4th 1178 (9th Cir. 2025), the Ninth Circuit determined that the Shelby GT-500 Ford Mustang codenamed "Eleanor" from the original and remake of "Gone in 60 Seconds" (and two other movies) was not entitled to character copyright protection. To complicate things, the rapid adoption of generative AI is introducing additional uncertainty into this already unsettled doctrine, raising difficult questions about whether preexisting characters that appear in AI-generated works will receive the same degree of copyright protection.

Existing Character Copyright Law

Federal copyright law lists various categories of protected subject matter, such as literary works, motion pictures, and pictorial and graphic works.[1] Although the statute is silent as to the protection of characters within these works, "there is a long history of extending copyright protection to graphically-depicted characters."[2] But "not every comic book, television, or motion picture character is entitled to copyright protection."[3] Although different courts apply different tests when assessing whether a character is entitled to copyright protection, courts—like the Ninth Circuit—will typically assess the following: (1) whether the character has "physical as well as conceptual qualities," (2) whether the character is "sufficiently delineated to be recognizable as the same character whenever it appears" and displays "consistent, identifiable character traits and attributes," and (3) whether the character is "especially distinctive" and contains "some unique elements of expression."[4]

So, in the case of the Eleanor Mustang mentioned above, the court looked at 11 iterations across four films, finding the vehicle lacked conceptual qualities because it had "no anthropomorphic traits"; it "never acts with agency or volition," as it is always "driven by the film's protagonists" [without any] "sentience, emotion, or personality." The car likewise lacked consistent traits because its "physical appearance changes frequently, appearing as yellow and black Fastback Mustang, a gray and black Shelby GT-500 Mustang, and a rusty, paintless Mustang in need of repair." Nor was it "especially distinctive from any number of sports cars appearing in car-centric action films." "In sum, Eleanor is not really a character."[5]

Generative AI and Copyright Uncertainty

The advent of generative AI adds new wrinkles to character copyright law. The Copyright Office and courts have found that purely AI-generated works—which would presumably include characters—are not copyrightable because they are considered machine authored, not human authored.[6] So if someone asks AI to generate an original feline-like creature with wings, that cat-like character would likely be unprotectable. But that's the easy question, at least for now. Much harder issues remain unresolved and need additional legal development.

For example, courts have historically treated well-defined characters as protectable independent of the specific works they appear in. However, courts will look at past depictions of the character to determine a character's physical and conceptual qualities and whether it is sufficiently delineated whenever it appears, with consistent, identifiable traits and attributes that make it distinctive. Under this logic, an AI-generated image or film may lack copyright protection, but a preexisting character that appears in that image or film would still be copyrightable. But could the use of protectable characters in generative AI works actually weaken the preexisting character's copyright? Because copyright protection for characters depends on consistent delineations across prior works, if rightsholders or others use AI to generate numerous variations, that character could become less consistently defined, which could then impact protection arguments.

Nor is it always clear what happens to an AI-generated work when a protectable character appears in it. In 2023, the U.S. Copyright Office denied copyright registration for a visual artist Ankit Sahni's copyrightable photo of a sunset that he fed into an AI tool and asked it to produce a variation with stylistic elements of Van Gogh's "The Starry Night." The Copyright Office refused to register the entire new "Van Goghified" image, finding that the new work lacked human authorship because the applicant "exerted insufficient creative control over [the tool's]" generation of the output.[7] That decision leaves characters copyright owners in a legal gray area, where (1) their underlying character may still be copyrightable; (2) the AI output featuring that character has no registrable authorship and is in the public domain; and (3) the human-authored character inside the AI output does not make the output registrable.

Or take this area of ambiguity for those interested in using AI to help create new characters: What happens if a human generates a character that is later conceptualized with AI? For example, if an artist writes a prompt of a cyberpunk detective character with neon tattoos and a robotic arm, is that prompt sufficient creative direction to make the AI generated character copyrightable? Or is the prompt just an unprotectable conceptual input, leaving the expression to the model—thereby making the character unprotectable? The Copyright Office has increasingly leaned towards the latter view, meaning prompt writers likely do not own the characters they generate with AI. But what happens if human authors later add additional, human-authored characteristics to the character? For instance, what if the artists draw new designs of the cyberpunk detective character, give him a soft spot for cats and ice cream, and feature him in a series of human-authored books (where he has consistent, identifiable attributes)? Can these human actions eventually imbue the character with enough human authorship to generate character copyrightability from a character originally conceptualized with AI? We may be years from finding out the answer to these questions.

So where does all of this uncertainty leave IP owners and content creators? For now, caution and documentation are likely the safest path forward. Rights holders experimenting with generative AI may want to treat AI character outputs as preliminary or assistive material rather than final expressive works—and may want to ensure that human creators meaningfully shape the final character depiction through significant editing, redrawing, or other creative contributions. As GenAI tools become more integrated into creative pipelines, courts and the U.S. Copyright Office will need to clarify how these technologies interact with longstanding character protection doctrines. But until then, creators may find themselves navigating a peculiar reality: preexisting characters that remain protectable in theory, but whose appearances in AI-generated works may occupy an uncertain legal space.

+++

Adrian Vallens is an associate in the intellectual property group in the Los Angeles office of DWT. For questions or more insights, reach out Adrian or another member of our intellectual property team and sign up for our alerts.