The copyright question haunts every designer who has touched an AI tool in their creative process. Can you own the output? Does using AI void your protection on the broader work? If your design was informed by an AI suggestion, where does AI end and your authorship begin?
The answers are clearer than the coverage suggests, but they require understanding the distinction between AI-generated work and AI-assisted work, and building a documentation practice that supports your ownership claims.
The foundational rule: human authorship required
The Copyright Office's March 2023 registration guidance established the current rule: copyright requires human authorship. Works generated entirely by AI, produced by entering a prompt and accepting the output without substantial human creative shaping, do not qualify for copyright protection.
The courts agree. In Thaler v. Perlmutter, decided in March 2025, the D.C. Circuit held that the Copyright Act requires an author to be a human being, rejecting Dr. Stephen Thaler's attempt to register a work autonomously created by his "Creativity Machine." The Supreme Court declined to take the case in 2026. That question is settled.
The Office went further in its January 2025 report, Copyright and Artificial Intelligence, Part 2: Copyrightability. Its conclusion on prompting is blunt: prompts alone do not make you an author. In the Office's words, prompts "essentially function as instructions that convey unprotectable ideas," because current tools don't give the user enough control over how the idea becomes expression.
This means: if you type a text prompt into Midjourney and use the resulting image as your campaign visual, you do not own the copyright in that image. Neither does Midjourney's developer, and neither does anyone else. As far as US copyright is concerned, a competitor who copies it owes you nothing. The same applies to patterns, prints, and designs generated by AI without substantial human creative input. "Make me a floral pattern in the style of William Morris" does not make you the author of the resulting pattern, regardless of how specific or creative your prompt was.
Where protection survives: AI-assisted work
The landscape becomes more interesting, and more valuable for working designers, when we move from AI-generated to AI-assisted work. This is the territory where most serious creative practice actually lives.
The Part 2 report identifies three ways human authorship can attach to a work made with AI: human-authored material that is perceptible in the output, creative selection, coordination, and arrangement of AI-generated material, and creative modifications of AI output. The AI-generated raw material stays unprotected. Your creative work on top of it does not.
Two registration decisions show the line in practice. In Zarya of the Dawn, a 2023 decision involving a graphic novel with Midjourney-generated images, the Office refused protection for the individual images but registered the human-written text and the author's selection and arrangement of the elements. And in January 2025, the Office registered A Single Piece of American Cheese, an image built through dozens of rounds of inpainting, selecting regions of an AI image and regenerating them, based on the human "selection, coordination, and arrangement of material generated by artificial intelligence." These are registration decisions, not court rulings, but they are the clearest map we have.
A designer who generates 50 AI pattern variations, then selects, arranges, scales, recolours, and integrates one specific option into a collection has made significant creative choices. Those choices, the selection, the arrangement, the modifications, the integration into a broader designed object, may be protectable even if the underlying AI output is not.
For fashion designers, this is the practical framework: the creative choices you make in working with AI tools, what you generate, what you reject, how you modify the output, how you combine it with other elements, how you apply it to a garment, are the authorship. Those choices need to be documented.
The litigation: where the money questions are being decided
The copyrightability rules above are relatively settled. The live fights are about training, and they will determine what rights you have when your runway photographs, lookbooks, and editorial imagery are used to train AI systems.
The bellwether visual arts case, Andersen v. Stability AI, brought by illustrators against Stability, Midjourney, DeviantArt, and Runway, survived motions to dismiss on core claims but ended in early 2026 without a trial or a merits ruling on the training question. In the text world, 2025 gave us split signals: courts found training on lawfully acquired books could be fair use, while Anthropic agreed to a settlement reported at $1.5 billion over its use of pirated books. The image cases that matter most to fashion are still unresolved. I keep a running map of every case in the litigation tracker, and the training-data question gets its own deep dive in Your Designs Are Probably in an AI Training Dataset.
The honest summary: whether AI training on your copyrighted imagery infringes, and whether AI output can infringe the works it was trained on, are questions courts have only begun to answer. Anyone who tells you the answer is settled is selling something.
The practical documentation framework
The designers who will successfully protect AI-assisted work are those who can demonstrate their human creative choices. The Copyright Office asks registration applicants to disclose AI-generated content in submitted works, and it looks for evidence that the human made genuine creative decisions, not just that they operated the tool.
Build this documentation practice now:
Save your creative brief before you open the AI tool. A written brief that describes what you're trying to achieve, what aesthetic direction you're working in, and what formal problem you're solving establishes that you came to the AI tool with a creative intention. You weren't passively accepting whatever it produced.
Document what you rejected and why. If you generated 40 pattern options and selected 3 for further development, keep the full generation and note why you chose what you chose. The act of selection is creative authorship. The evidence of selection supports your claim.
Record every modification you made to AI output. Colour adjustments, scale changes, compositional modifications, combination with other elements. Under the Part 2 report, creative modifications are one of the recognised paths to protection, and they are the clearest evidence of human authorship. Provenance metadata can capture much of this automatically; see our provenance guide.
Photograph your design development process. Sketches, material samples, iteration photographs, pattern development, anything that shows the human creative process that surrounded and shaped your use of AI tools.
Register your copyrights. Registration is not required to own a copyright, but for US works it is required before you can file an infringement suit, and timely registration is what unlocks statutory damages and attorney's fees. Registration also creates a public record with a timestamp. When you register a work containing AI-generated material, disclose it as the Office's guidance requires. For significant commercial works, it's inexpensive insurance.
What about your designs that are already in AI training data?
Every image you've published publicly, every campaign, every runway photograph, every editorial, has quite possibly been swept into AI training datasets. Whether that constitutes infringement is the question the courts are still working through. We cover what you can do about it, including California's new training-data disclosure law, in Your Designs Are Probably in an AI Training Dataset.
This article is editorial analysis, not legal advice. For questions about your specific situation, consult a qualified attorney.