AI Image Copyright: Can You Own AI-Generated Images?

AI image copyright in the US is clear on one point. Under current Copyright Office guidance, purely AI-generated images are not protected, and a prompt alone usually does not give you enough control to claim authorship. Human contributions, such as creative selection, arrangement and substantial edits, can be protected. The rules differ in the UK, the EU and China. This is general information, not legal advice.

The US rule: human authorship

The US Copyright Office published Part 2 of its report on copyright and artificial intelligence, covering copyrightability, in January 2025. Its report page and the PDF set out the conclusions. In summary:

Copyright protects original human expression, even when a work includes AI-generated material.
It does not extend to purely AI-generated material, or to material where humans lack control over the expressive elements.
Humans can claim their perceptible contributions, their creative selection, coordination or arrangement, and their creative modifications.
Using AI as an assistant does not by itself reduce protection for the human parts.
"Based on the functioning of current generally available technology, prompts do not alone provide sufficient control."
Repeatedly revising a prompt "does not change this analysis". The Office compares it to "re-rolling the dice".

The courts have backed the human-authorship requirement. In Thaler v. Perlmutter, the D.C. Circuit held on 18 March 2025 that the Copyright Act requires a human author. The Supreme Court denied certiorari on 2 March 2026, according to the Supreme Court docket summary. A denial of certiorari is not a ruling on the merits. It also does not settle how much human involvement is enough for an AI-assisted work.

What counts as enough human input

The Office's report says that tools letting a user select and regenerate regions, and edit the output, can give control over "the selection and placement of individual creative elements". Whether that is enough is decided case by case. The report uses Midjourney's Vary Region feature as an example.

Two other points from the report matter for images:

Selecting and arranging only two or three elements is generally not enough for protection.
In the Zarya of the Dawn decision (21 February 2023), the Office registered the human-written text and the human selection and arrangement of text and images. It did not cover the Midjourney images themselves.

Registering an AI-assisted image

If your work contains more than a de minimis amount of AI-generated material, the Office's registration guidance (88 Fed. Reg. 16,190, 16 March 2023) asks you to disclose it and briefly describe your own contribution. Registration then covers the human contribution, not the AI material. The Office says it has registered hundreds of works that include AI material on that basis.

In practice, keep a record of what you did: the prompts, the edits, the regions you selected, and the arrangement choices. That record makes it easier to describe your contribution accurately.

Cases to watch

Three cases are still moving, and none has settled the questions most creators ask.

Allen v. Perlmutter (US, District of Colorado): the case challenges the Office's refusal to register a Midjourney image made with more than 600 iterative prompts. Both sides have moved for summary judgment. We could find no ruling as of October 2026, so check the docket before you rely on the position.
Getty Images v Stability AI (UK): the High Court handed down its judgment on 4 November 2025, [2025] EWHC 2863 (Ch). Getty's remaining secondary copyright claims were dismissed, and some trade mark claims succeeded in a way the judge described as "extremely limited in scope". Permission to appeal on the secondary-infringement point was reported in December 2025. We found no hearing date.
Disney and Universal v Midjourney (US): filed in June 2025, with a similar suit from Warner Bros. Discovery in September 2025. The cases were combined in November 2025, and the judge suggested mediation. We found no ruling. These cases are about training and outputs, not about who owns your images.

AI image copyright by country

| Jurisdiction | Purely AI-generated image | AI-assisted image | Status, October 2026 | |---|---|---|---| | United States | Not protected under current Office guidance | Human contribution can be protected, case by case | Office guidance (January 2025); Thaler ruling; certiorari denied March 2026 | | United Kingdom | Computer-generated works have been protected since 1988; the government proposes removing that protection | AI-assisted works would keep protection under the proposal | Proposal in the government report of 18 March 2026, not enacted law | | European Union | No binding ownership rule. Parliament's non-binding resolution says fully AI-generated content should not be protected | Protection rests on human authorship | Resolution adopted 10 March 2026 | | China | A Beijing court protected a heavily directed image and treated the user as the author | The human user is the author | First-instance ruling, 27 November 2023 |

The China ruling is Li v. Liu. The Beijing Internet Court found that the user, who chose more than 150 prompts, arranged them, set parameters and kept refining the image, was the author. The AI model was not. It is a first-instance decision, and Chinese courts do not follow binding precedent, so it does not settle the question elsewhere. A Wolters Kluwer analysis notes that it contrasts with US registration decisions.

The EU position is less settled. Parliament's resolution is available from the European Parliament, and it calls for clear labelling of purely AI-generated content. It does not create binding rules on ownership. The UK government's March 2026 report proposes removing the protection for computer-generated works with no human author, but that is a proposal.

Who owns the output under tool terms?

A tool's terms can say what you may do with an output. They do not decide whether copyright exists. OpenAI's terms, for example, assign whatever output rights exist, to the extent the law allows. Check each tool's terms. Assigning whatever rights exist is not the same as the output being protected by copyright.

Google's generative AI terms page that we checked is marked as no longer applying from 22 May 2024, and it does not address output ownership. Check the terms that apply to the product you use.

Adobe says Firefly outputs are "safe for commercial use", based on licensed and public-domain training content. That statement is about training data and commercial risk. It does not say whether you can copyright an output.

Using AI images commercially: a practical checklist

Confirm the plan allows commercial use. That permission is a licence from the tool. It is separate from whether copyright exists in the output.
Keep records. Save the prompts, the edits and the choices you made about selection and arrangement.
Document your human contribution. Note what you drew, painted, composed, selected or changed.
Read the output terms. Check who the terms say owns the output, and any limits on use.
Check for real people and trademarks. Look at the output for recognisable people and brand marks before you publish.
Disclose AI use where required. Platform rules and contracts may require it.
Get advice for high-value work. A lawyer can apply the rules to your facts.

Provenance and labelling

Provenance tools can show where an image came from, though they are not proof of authorship. Google says all Gemini-generated images include a SynthID watermark, and it pairs SynthID with C2PA Content Credentials. OpenAI says its images carry C2PA metadata and a SynthID watermark, and warns that metadata can be removed. Labels help people understand an image's origin. They do not decide who owns it.

Where OndaLabs fits

OndaLabs includes commercial use on its Starter plan ($9 per month) and above, and image generation is included from Starter. That is a permission from OndaLabs to use outputs in business under the plan. It is separate from copyright. OndaLabs does not own, assign or indemnify the images you make, and we do not claim the images are copyright-safe, licensed-trained or watermarked. You can read plan details on the pricing page and try the image studio.

For related questions about music, read our AI music copyright guide. For the prompting side of image work, see how to write AI image prompts that work.

Frequently asked questions

Can you copyright an AI-generated image?

Not if it is purely AI-generated, under current US Copyright Office guidance. Human contributions, such as selection, arrangement and creative edits, can be protected, and the answer depends on the facts. This is general information, not legal advice.

Does writing a detailed prompt make me the author?

The US Copyright Office says prompts alone do not provide sufficient control with current technology, and repeatedly revising a prompt does not change that analysis. A Beijing court reached a different result on its facts. Allen v. Perlmutter, which turns on this question, has no ruling that we could find.

Can I register an AI-assisted image with the US Copyright Office?

Yes, if you disclose the AI-generated material and describe your own contribution. The Office's 2023 guidance asks for that disclosure when the AI material is more than de minimis. Registration covers your contribution, not the AI material.

Is an AI image safe to use commercially?

It depends on the tool's terms and on the content itself. A plan's commercial-use permission is a licence from the tool. It does not mean the image is copyrightable, and it does not remove the need to check for recognisable people or trademarks.

Are the rules the same in the UK, the EU and China?

No. The UK protects computer-generated works today, but the government proposes removing that protection. The EU has a non-binding Parliament resolution and no binding ownership rule. China has a first-instance ruling that protected a heavily directed image. Check the rules for your jurisdiction.