Made With or Made By
The Unavoidable Copyright Question of AI use

A photographer’s layperson perspective

The author and editors work primarily in fine art photography in the U.S. This perspective is shaped by that context.
This article is for informational purposes only and does not constitute legal advice. For advice on your situation, consult a qualified attorney.

From a federal courtroom to post processing with multiple rounds of inpainting, copyright law is drawing lines through territory most fine art photographers haven't mapped yet.

In 2023, author Peter von Stackelberg predicted that questions around AI-generated art and copyright law would “become increasingly complicated.” That turned out to be exactly right, just not in the direction most people assumed.

The legal system has spent the past several years answering one question loudly and clearly. The answers to the more interesting questions are only now coming into focus. And for fine art photographers working with AI tools, those answers matter more than the headlines suggest.

The settled question, and why it is not your question

On March 2, 2026, the United States Supreme Court declined to hear Thaler v. Perlmutter, the case that has dominated AI copyright discussion since 2018. Computer scientist Stephen Thaler had applied for a copyright covering “A Recent Entrance to Paradise,” a piece of visual art he said his AI system, DABUS, created entirely on its own. He listed the AI as the sole author. The Copyright Office rejected the application. Federal courts agreed. The DC Circuit affirmed in 2025. The Supreme Court left that ruling in place without comment.

The ruling is settled as a matter of current law: a machine cannot hold a copyright. An AI system, no matter how sophisticated, cannot be named as an author. No court has ever recognized copyright in material created by non-humans, and the ones that have engaged with the question directly have refused to.

Here is the part worth noting if you are an AI-assisted photographer: Thaler’s case was deliberately extreme. He explicitly described his AI as the sole author and stated the work was created without human intervention. He was not working with the tool. He stepped back and let it run, then tried to own the result. That argument was never going to succeed under existing law, and the dismissal of it does not say much about how copyright applies to the work most serious AI artists actually make.

The conversation that actually matters begins where Thaler’s ends.

The question the courts have not answered yet

The Copyright Office’s January 2025 report, its most comprehensive statement on AI and authorship to date, is careful to distinguish between two very different situations. Works generated entirely by AI, with no meaningful human creative input, cannot be copyrighted. But the report is equally clear that using AI as a tool does not automatically strip a work of copyright protection. The human-authored portions of a hybrid work remain protectable. The creative decisions made by a human throughout the process can constitute authorship.

What the report does not provide is a bright line. What it offers instead is a framework, and a pending case that will test it.

Jason Allen’s Théâtre D’Opéra Spatial won a prize at the Colorado State Fair in 2022, generating headlines about whether AI had “won” an art competition. Allen subsequently filed for copyright registration, disclosing his use of AI tools. The Copyright Office declined. Allen sued, arguing that his contribution involved more than 600 iterative prompts, extensive refinement choices, and significant curation of outputs. As of early 2026, that case remains active in federal court.

The core question in Allen’s case is the same question every fine art photographer using AI tools needs to think about: at what point does human creative direction become sufficient to constitute authorship of the final output?

What the Copyright Office actually considers authorship

The 2025 report draws the line not at whether AI was involved, but at whether the human exercised meaningful creative control over the expressive elements of the work.

Prompting alone, in the Office’s current view, does not constitute authorship. The reasoning is direct: identical or similar prompts produce meaningfully different outputs across different runs. If the user cannot determine the expressive content of the image through the prompt, the user is not controlling what the image looks like in any authorial sense. The Office describes iterative prompting as akin to repeatedly requesting something from a commissioned artist rather than making the creative decisions yourself.

The registration that currently marks the outer boundary of what the Office will protect is a work called A Single Piece of American Cheese, registered in early 2025. The artist used Invoke AI’s inpainting tools across 35 separate rounds of iterative modification. The registration turned not on the prompts themselves but on the applicant’s demonstrated human authorship in the selection, coordination, and ongoing refinement of each successive state of the image.

That framing will be immediately recognizable to photographers who work extensively in post-production. The distinction the Office is drawing is essentially the same one that separates a photographer who selects a stock image from one who makes dozens of deliberate darkroom decisions about how a final print will look. Both involve tools. Only one involves authorial choices that shape the expressive content of the result.

What this means for a body of AI fine art work

Working through the implications for a sustained photographic practice with AI is worth doing in plain terms, because the legal picture is more favorable than the headlines imply, and more demanding than some practitioners want to acknowledge.

If you are using AI generation as one layer among many, where the output is a starting point that you subsequently direct, modify, composite, or reconstruct through human creative choices, those contributions are protectable. The human-authored elements remain yours. The AI-generated substrate, on its own, is not.

If you are generating an output from a prompt and presenting it without further intervention, the copyright position is weak under current guidance, regardless of how much craft went into the prompt. The Office has been explicit that prompting, as presently understood, does not give the user sufficient control over the expressive result to constitute authorship.

The case-by-case analysis the Office applies looks at whether a human author has made sufficient and demonstrable creative choices that determine the expressive content of the final work. The more of those choices you can document and articulate, the stronger the position.

Documentation, in this environment, is not bureaucratic excess. It is part of the practice.

The question nobody is asking about training data

Parallel to the authorship question is a dispute most practicing artists have less direct exposure to: whether the companies whose models they rely on are themselves infringing copyright in the training data those models are built on. This will shape the landscape in ways that matter.

In November 2025, the UK High Court delivered judgment in Getty Images v. Stability AI. The court found in Getty’s favor on trademark infringement but dismissed the secondary copyright claim, holding that Getty had not established that its specific images were used in the training process in the required way. The evidential bar for proving which specific works were used in a large-scale dataset proved difficult to clear.

In the United States, the Copyright Office’s Part 3 report, which addresses training data and fair use, was released in pre-publication form in May 2025. Its conclusion was that outcomes will be fact-specific: some uses of copyrighted works for AI training will qualify as fair use, and some will not. Courts in related cases are actively working through those analyses.

The practical implication for AI-assisted photographers is not that the tools you are using are legally clean. It is that the resolution of these disputes will affect what tools remain available, what licensing arrangements develop, and what the commercial environment for AI-assisted work looks like over the next few years. Watching this space is not optional for anyone building a practice around these tools.

The longer view

Copyright law has always been uncomfortable with technology that blurs the line between authorial choice and mechanical process. The same discomfort that now surrounds AI prompting appeared when courts first had to decide whether photographs were eligible for copyright protection at all. The question then was whether pointing a lens at something and pressing a shutter constituted original authorship, or whether a machine was simply recording what was already there.

The answer, eventually, was that what photographers do with light, with framing, with timing, and with the choices made before and after the exposure constitutes authorship, even though the final result is produced by a machine. The creative decisions matter. The tool does not disqualify the maker.

That precedent does not automatically extend to AI-generated work in a tidy way. The legal architecture is different, the creative process is different, and courts will need to work through those differences on their own schedule. But the underlying logic is not alien territory for photography. The courts have been here before.

What the current moment requires of AI-assisted fine art photographers is clarity about where in the process the human creative decisions actually live, and the discipline to be able to account for them. Not because the law demands a performance of effort, but because authorship has always required the ability to locate the maker’s hand in the work.

If you can do that honestly, the copyright question is more navigable than the current headlines suggest. If you cannot, that is a question worth sitting with, whatever the legal implications turn out to be.