Icon

Updated weekly

OW

OW

Icon

Updated weekly

OW

Output Weekly

Output Weekly

Tools, workflows and honest rankings for people who make things with AI

Tools, workflows and honest rankings for people who make things with AI

Platforms

Can you copyright AI-generated images?

Sam Whitlock

Rows of old leather-bound books on library shelves

Key takeaways

  • In the US, an image generated from a text prompt alone can't be copyrighted, because the Copyright Office says prompts don't give enough control over the result.

  • The courts back the human authorship rule. The D.C. Circuit affirmed it in Thaler v. Perlmutter in March 2025, and the Supreme Court declined to hear the case on March 2, 2026.

  • Your own contributions can be protected, including artwork that stays visible in the output, creative edits, and the arrangement of images in a larger work.

  • To register a work with AI material in it, you disclose that material and exclude anything more than a trivial amount. The Office has registered more than 7,000 claims that way.

  • Other countries differ. UK law still protects "computer-generated" works, though the government proposed removing that in March 2026. A Beijing court granted copyright to a Stable Diffusion image in 2023.

Can you copyright an AI-generated image?

Not the AI-generated part, at least in the US. An image a model produces from your text prompt is not protected by copyright, however long you spent on the prompt, because US law requires a human author and the Copyright Office says the model, not the prompter, decided what the image looks like.

The human parts of a work made with AI can still be protected, such as a drawing you fed in that still shows in the output, creative edits, or the way you arranged many images into a comic or a book. This piece is general information, not legal advice.

What the Copyright Office says

The Office's main statement is Part 2 of its report on copyright and AI, published on January 29, 2025. It concludes that existing law can handle these questions with no new legislation, and it sets out a handful of principles.

  • Using AI to assist a human creator doesn't affect copyright in what that person made.

  • Copyright doesn't cover purely AI-generated material, or material where there's "insufficient human control over the expressive elements."

  • Whether a human contribution is enough has to be decided case by case.

  • "Based on the functioning of current generally available technology, prompts do not alone provide sufficient control."

The report compares prompts to instructions given to a commissioned artist. They describe what you want, but the system decides how to carry them out. Revising the prompt many times is just "re-rolling" the dice, in the Office's words, which gives you more outputs to pick from but no more control over any one of them. The Office says it may revisit these conclusions as the technology changes.

The rules for applicants come from the Office's March 2023 registration guidance, still the current version. In May 2026, Register of Copyrights Shira Perlmutter told a Senate subcommittee that more guidance on disclaimers was coming later in the year, but none had been posted by early November.

Thaler v. Perlmutter and the human authorship rule

Stephen Thaler listed his AI system, the Creativity Machine, as the sole author of an image called "A Recent Entrance to Paradise." The Office refused to register it, and the district court and the D.C. Circuit agreed. In its March 18, 2025, opinion, the appeals court held that the Copyright Act "requires all eligible work to be authored in the first instance by a human being."

The same opinion said the rule "does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence," as long as the author is a human. The Supreme Court denied review on March 2, 2026, so the ruling stands. The case left the harder question open, since Thaler never claimed any creative contribution of his own.

Zarya of the Dawn and AI art in a comic book

Kristina Kashtanova's 18-page comic book was registered in 2022 before the Office learned its images came from Midjourney. In a February 2023 letter, the Office narrowed the registration to her text and "the selection, coordination, and arrangement of the Work's written and visual elements," leaving out the Midjourney images. It called her edits to one character's lips "too minor and imperceptible" to count.

Théâtre D'opéra Spatial and the Allen case

Jason Allen's image won the 2022 Colorado State Fair's fine art competition. He told the Office he entered prompts and revisions "at least 624 times" in Midjourney, then edited the result in Photoshop and upscaled it with Gigapixel AI. The Office said it would register his human-made Photoshop edits, as long as the Midjourney material was disclaimed. Allen refused, and the Review Board upheld the refusal in September 2023.

Allen sued in federal court in Colorado in 2024. The government's cross-motion for summary judgment, filed February 6, 2026, argues that his prompts couldn't control the image's expression. As of early November 2026 the court hadn't ruled, and the case asks directly whether heavy, iterative prompting can make someone an author.

Rose Enigma, SURYAST, and your own art as input

The artist Kris Kashtanova fed a hand-drawn illustration into an AI system with a prompt and registered the result, Rose Enigma, in 2023. The registration covers only the human drawing that's "clearly perceptible" in the output, such as the outline of a mask and the placement of leaves.

SURYAST went the other way. Ankit Sahni put a photograph he'd taken into an AI app and had it restyled after Van Gogh's "The Starry Night." In December 2023 the Review Board refused registration, finding he had too little control over an output that didn't clearly show his photo.

What counts as enough human input

Putting the report and the decisions together, common workflows map onto the rules roughly like this.

What you did

Likely treatment in the US

Wrote a prompt, even a long one, and picked the best result

Not protected

Re-prompted many times until one output looked right

Not protected, per the Office ("re-rolling")

Fed in your own drawing or photo, and it's still visible in the output

Your visible expression is protected, but the AI additions aren't

Edited the output substantially by hand or in an editor

Your edits can be protected if they're creative enough

Selected and regenerated specific regions (inpainting)

Possible, decided case by case

Selected and arranged many images into a comic, book, or series

The selection and arrangement can be protected

Made a small fix, such as a minor touch-up or an upscale

Usually not enough

On inpainting, the report uses Midjourney's Vary Region tool as an example and says such tools, "unlike prompts alone," can let the user control "the selection and placement of individual creative elements." Whether that clears the bar depends on the facts.

A tool that steers generation with your own sketch, such as a ControlNet reading line art you drew, is closer to Rose Enigma than a text prompt is, though the Office judges what's perceptible in the final image, not the tool. Training a LoRA on your own artwork isn't addressed in the report or the decisions above. AI material inside a bigger human work doesn't cost that work its protection, either. The report says a film with AI-generated effects is still copyrightable as a whole, though the effects themselves aren't.

How registration works for AI-assisted work

The 2023 guidance says applicants "have a duty to disclose the inclusion of AI-generated content" and to briefly explain what the human author contributed. It spells out the steps.

  • Use the Standard Application, and describe your own contribution in the "Author Created" field.

  • Exclude AI-generated material that's more than de minimis under "Material Excluded" in the Limitation of the Claim section.

  • Don't list the AI tool or its company as an author or co-author.

  • If you're unsure, say the work contains AI-generated material, and the Office will follow up.

A registration that didn't disclose AI material should be fixed with a supplementary registration. The guidance warns that applicants who leave the record uncorrected "risk losing the benefits of the registration." Disclosure isn't a dead end, either. Perlmutter's May testimony said the Office had registered more than 7,000 claims that include AI-generated material, with that material disclaimed.

What this means for selling and licensing AI images

You can sell or license an AI-generated image. The catch is that you can't stop other people from copying what you don't own. If a print or a book cover is purely AI-generated, a competitor who copies it hasn't infringed any US copyright, because none exists. That has a few consequences.

  • A generator's terms of service may say you own your outputs. That sorts out rights between you and the company, but it doesn't make the image copyrightable.

  • If a buyer wants exclusive rights, be clear about which parts you actually hold rights in. Your own drawing, edits, or arrangement may be protected even when the base image isn't.

  • To sue over a US work, you generally need a registration first, under 17 U.S.C. § 411(a), and statutory damages depend on registering in time. For AI-assisted work, that means filing with an accurate disclaimer.

  • Ownership and infringement are separate questions. An image you can't copyright can still infringe someone else's.

How other countries handle it

The UK is the best-known exception. Its Copyright, Designs and Patents Act 1988 protects "computer-generated" works, meaning works with no human author. The author is the person who made "the arrangements necessary for the creation of the work," and protection lasts 50 years. In its March 18, 2026 report on copyright and AI, the government said that "in the absence of evidence of its ongoing value, we propose that this specific type of protection should be removed," while keeping protection for works made with AI assistance. As of November 2026, the provision is still in force.

China's best-known case went the other way from the US. In Li v. Liu, decided on November 27, 2023, the Beijing Internet Court found that an image made with Stable Diffusion was protected, and that the person who generated it was the author. The court pointed to his choice of prompts and parameters and his later adjustments, which it said reflected "aesthetic choice and personal judgment," and awarded him 500 yuan. The Copyright Office's report notes that the ruling isn't precedential in China.

Elsewhere, the Copyright Office's report finds broad agreement that copyright needs a human author. South Korea's 2023 guidance says an AI output can't be registered without a creative human contribution. Japan's 2024 guidelines weigh the prompts, the number of attempts, the choice among outputs, and later human edits case by case. And in a 2024 Council of the EU questionnaire, member states said AI output should be eligible only if the human input was significant.

Copyright and AI images FAQ

Can I copyright an image I made with Midjourney or another generator?

Not the generated image itself, if all you did was write prompts. You can claim your own edits, artwork, or arrangement, as long as you disclose and exclude the AI material.

Does writing a very detailed prompt make me the author?

No, according to the Copyright Office, and revising the prompt many times doesn't change that. A pending case, Allen v. Perlmutter, is testing that position in federal court.

Did the Supreme Court rule on AI copyright?

No. On March 2, 2026, it declined to hear Thaler v. Perlmutter, which leaves in place the D.C. Circuit's ruling that a machine can't be the author of a copyrighted work. The Court hasn't addressed how much human involvement is enough.

Do I have to tell the Copyright Office I used AI?

Yes, when you apply to register the work. The Office's 2023 guidance requires you to disclose AI-generated material and exclude anything more than de minimis from your claim. Registrations that left it out should be corrected with a supplementary registration.

Can I sell AI-generated images?

Yes. Copyright doesn't stop you from selling them, but it also won't stop anyone from copying the purely AI-generated parts. Check the generator's terms of service for any limits on commercial use.