Style Mimicry and Copyright: Where the Line Is

TL;DR: Most AI art copyright confusion comes from blurring two separate questions. Style itself is generally an unprotected "idea" under the statute, but "style mimicry" also raises whether you can copyright the AI output, and whether reproducing an artist's actual expression (or naming them to imply endorsement) creates liability. Courts have decided pieces of this; the hardest questions are still open litigation. Not legal advice.

What copyright actually protects, and what it doesn't

The starting point isn't a court case at all. It's the statute. U.S. copyright law states its own limits directly, in 17 U.S.C. § 102(b): "In no case does copyright protection for an original work of authorship extend to any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work."

The U.S. Copyright Office's own consumer-facing FAQ restates this plainly: "Copyright does not protect ideas, concepts, systems, or methods of doing something. You may express your ideas in writing or drawings and claim copyright in your description, but be aware that copyright will not protect the idea itself as revealed in your written or artistic work."

An artistic style, brushwork, a characteristic palette, a way of composing a scene, is generally analyzed the same way: as a method or approach, not as the specific expression the statute protects. That's why "paint in the style of Impressionism" has never been a copyright question. What changes with a named living artist and an AI tool is not the doctrine; it's that the output can get close enough, fast enough, to raise two adjacent questions the style-is-an-idea rule doesn't answer by itself.

Question one: can you even copyright the result?

This is settled further than people assume, and it cuts against the AI user, not in their favor. On March 18, 2025, the D.C. Circuit decided Thaler v. Perlmutter, affirming the Copyright Office's refusal to register an image whose sole listed author was an AI system. The court's holding was direct: "The Creativity Machine cannot be the recognized author of a copyrighted work because the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being."

The more useful case for style specifically is one the Copyright Office decided two years earlier, and it's about exactly this scenario. In December 2023, the Office's Review Board considered "SURYAST": a work made by feeding an original photograph and a copy of Van Gogh's The Starry Night into an AI tool called RAGHAV as the "style" input, then choosing a numerical value for how strongly the style should transfer. The applicant argued that his choices, the original photo, the selection of that particular painting as style reference, and the strength setting, added up to enough creative control to make him the author. The Board disagreed: "RAGHAV’s interpretation of Mr. Sahni’s photograph in the style of another painting is a function of how the model works and the images on which it was trained on—not specific contributions or instructions received from Mr. Sahni." It concluded that "selecting a single number for a style filter is the kind of de minimis authorship not protected by copyright."

Read together with the Copyright Office's Part 2 report on copyrightability (released Jan. 29, 2025), the rule is consistent: the report's own conclusions state that "copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements." Typing "in the style of X" and picking a strength slider doesn't clear that bar. Drawing your own composition and using AI only to render it in a described style is a different, stronger case, one the Office's report treats separately, where a human's own perceptible creative choices remain protected.

Question two: does mimicking someone's style infringe their copyright?

This is the question the internet argues about, and it's where the honest answer is: it depends on what "mimicking" actually did, and courts haven't finished deciding the hardest version of it.

Two live cases illustrate the split, and neither has produced a final verdict resolving the core dispute.

Neither case stands for "AI training infringes copyright" or "AI training is legal," full stop, and this post won't predict which way the open U.S. theories land. What the Getty judgment shows is that the theory Getty actually litigated (a specific UK secondary-infringement statute) failed, on facts specific to that theory and that jurisdiction, after the broader training claim was withdrawn rather than decided. What Andersen shows is that a differently-framed theory, that specific training images are reproduced inside the model itself, is still being tested on the merits in a separate case, in a separate jurisdiction, on separate law. If you read a confident claim that "courts have ruled AI training is [legal/illegal]," ask which case and which specific claim, because the honest picture right now is unresolved, not settled either way.

There's a separate, still-pending piece from the regulator rather than a court: the Copyright Office's own Part 3 report, on whether training generative AI on copyrighted works is fair use, exists only as a pre-publication version dated May 9, 2025. As of this check, no final version has been published. Even the government's own analysis of the core question is, formally, not finished.

The question people actually mean: naming a living artist

Separate from all of the above, and often what "style mimicry" discussions are really worried about: typing a specific living artist's name into a prompt to get output that reads as theirs.

This isn't cleanly a copyright question at all. If style itself isn't protected expression, naming the artist doesn't change that analysis on its own. What it can implicate is different law entirely, the kind that governs using someone's identity to imply their involvement or endorsement. Andersen v. Stability AI carries exactly this theory against Midjourney: plaintiffs alleged Midjourney circulated a promotional "Names List" of artists and a "showcase" of user work incorporating those names. In an order dated Aug. 12, 2024, the court declined to dismiss the claim, holding that "whether or not a reasonably prudent consumer would be confused or misled by the Names List and showcase to conclude that the included artists were endorsing the Midjourney product can be tested at summary judgment." That's trademark-adjacent law about implied endorsement, not copyright, so a copyright defense like fair use has no bearing on it either way.

Practically, this means the riskiest habit isn't describing a palette or a brushstroke technique. It's outputting and distributing work that reads to a viewer as "a new piece by [named living artist]," especially in a commercial context where it could be mistaken for their actual output or an endorsement. That's a different exposure than a copyright claim, and it doesn't disappear just because the underlying style isn't itself protectable.

What the tools themselves say

Vendor terms are not a substitute for law, but they're worth reading because they tell you what the company itself is and isn't willing to promise.

Midjourney's Terms of Service (version effective date May 27, 2026, as published on its own documentation site) disclaim any warranty that generated Assets don't infringe: both the Services and the Assets are provided "WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE." The same document tells users directly: "Please consult Your own lawyer if You want more information about the state of current intellectual property law in Your jurisdiction." That's Midjourney's own terms admitting the ground is unsettled enough that it won't offer a legal opinion, and neither should this post pretend to on its behalf.

OpenAI's Usage Policies (effective Oct. 29, 2025, checked directly against the live document) contain no clause specific to artistic style or named artists at all. The closest relevant language sits under a general "Protect people" prohibition on using the service for anything that would "infringe on intellectual property rights of others," alongside an unrelated restriction on using "someone’s likeness, including their photorealistic image or voice, without their consent in ways that could confuse authenticity." Neither line addresses style by name; both are broad enough to reach the named-artist-endorsement scenario above without saying so explicitly.

A practical, hedged way to think about it

None of this adds up to a bright line, because there isn't one yet. What follows is a practical lean, not legal advice, and it should be read as one.

Describing visual characteristics, palette, brushwork, era, lighting, composition, sits on the "idea" side of the statute that has never been controversial, with or without AI. It's also how two companion pieces on this site, on typography prompting with Ideogram 4.0 and on keeping one visual style across a whole series, both recommend working, for the same underlying reason: description is durable and portable, and a name is a shortcut that carries risk description doesn't.

Naming a specific living artist, especially for anything commercial or anything distributed publicly, adds a distinct kind of exposure that has nothing to do with whether "style" is copyrightable: it can read as implying that person made or endorsed the work, which is the theory Andersen actually presses against Midjourney, separate from copyright entirely. Naming a school, movement, or era ("art nouveau," "Ashcan school," "Dutch Golden Age still life") avoids that specific problem because no living, identifiable person's endorsement is implied.

And regardless of what you named, reproducing recognizable, specific protected elements, an exact composition, a distinctive character design, a specific copyrighted image close enough to be substantially similar, is a different and older kind of copyright risk that predates generative AI entirely and that no amount of "it's just style" reframing changes.

Describe the STYLE, not the artist:
"[MEDIUM], [PALETTE], [ERA or MOVEMENT], [COMPOSITION APPROACH], [LIGHTING QUALITY]"

Example:
"Oil painting, muted earth-tone palette, early 20th-century Ashcan School realism,
tight urban composition with figures cropped at the frame edge, warm gaslight glow."

Avoid naming a living artist if the output is commercial, distributed publicly,
or could be read as implying that person made or endorsed it.

If you're building a prompt library around a recurring visual approach for a brand or a series, that's exactly the kind of reusable description Prompt Architects is built to hold: save the style description once as a template, keep it worded around the visual characteristics rather than a name, and reuse it across every generation instead of retyping (or drifting from) the same description each time. That's a workflow fix. It is not, and this article is not, a legal opinion on your specific use case.