Who owns the copyright of AI-generated images?

Short answer first: in the United States, an image generated purely from a text prompt has no human author, so it is not protected by copyright and no one owns it. You can claim copyright only on the parts a human actually authored: the selection and arrangement, and any substantial editing you add on top. The raw render itself sits outside copyright entirely.

That is a strange result the first time you meet it. You typed the prompt, you paid for the tool, the picture would not exist without you, and yet the law says the thing you made is not yours to own. The rest of this piece walks through why, what the courts actually decided, and where the line between "yours" and "no one's" really falls.

Can AI-generated images be copyrighted?

Not on their own. The U.S. Copyright Office position, set out in its 2025 report Copyright and Artificial Intelligence, Part 2: Copyrightability, is that copyright protects only work that is the product of human creativity. A purely prompt-generated image fails that test, because the human gives an instruction and the machine makes the expressive choices about how that instruction becomes an image.

The Office was specific on the point creators most want to be wrong about: prompts do not count as authorship. However detailed the prompt, however many times you refine it, the text is an instruction, not the authored work. That position is tied to the tools in use today rather than framed as a permanent rule, but it is where the Office and the courts stand now. So the default answer to "can AI images be copyrighted" is no, with an important exception covered further down.

Who owns an image you make from a text prompt?

Nobody, in the copyright sense. This is the part that trips people up, because there are two different things called "ownership" here and they have come apart.

You own the file. It is on your drive, you can post it, print it, put it in an ad, sell it. None of that requires a copyright. What you do not have is the exclusive right that copyright grants: the right to stop other people from copying it. If a prompt-generated image has no human author, there is no copyright in it, so there is nothing to enforce. Someone else can take that same image and reuse it, and you have no copyright claim to bring against them. In practical terms a pure render behaves less like property you own and more like something that fell into the public domain the moment it was made.

Copyright is not the only thing that can attach to an image. The terms of service of the tool you used can still set conditions on how you use the output, and that is worth reading. But those are private contract terms, not copyright, and they do not give you the right to stop other people from copying the image.

What did the courts actually rule?

The headline case is Thaler v. Perlmutter. Stephen Thaler tried to register an image made by an AI system he built, listing the machine as the author. The U.S. Court of Appeals for the D.C. Circuit affirmed the refusal in March 2025, holding that the Copyright Act of 1976 requires a work to be authored, in the first instance, by a human being. A machine cannot be the author.

The case did not end there, and the update matters: in March 2026 the Supreme Court declined to review the decision, letting the D.C. Circuit ruling stand. So the human-authorship requirement is settled at that level. What the courts have not resolved is the harder, more useful question for working creators, which is exactly how much human input it takes to cross from "machine-generated" into "protectable." That line is being drawn case by case, and that is where the exception lives.

When can you claim copyright on AI work?

When a human does something creative beyond prompting. The Copyright Office has registered works that contain AI-generated material, as long as the registration claims the human contribution rather than the raw output.

Three kinds of human authorship can carry a copyright here:

• Selection and arrangement. If you generate many images and curate, sequence and compose them into something with its own creative structure, that selection and arrangement can be protected, much like the editor of a photo book holds rights in the book without owning every photograph in it.
• Substantial editing. Meaningful human changes to an AI image, the kind that add your own expression rather than minor touch-ups, can support a claim in those changes.
• AI as one tool among many. Where AI assists a largely human-made work rather than standing in for the author, the human authorship is intact and protectable.

A useful reference point is Zarya of the Dawn, a comic with AI-generated panels. The Copyright Office registered the human-authored selection and arrangement of the comic and its text, but not the individual AI-generated images inside it. Same work, two answers, split exactly along the human-authorship line. One more practical rule follows from this: at registration you have to disclose the AI-generated content and describe what the human author contributed. Claiming the whole thing as your original work is how a registration gets invalidated.

Is it different outside the US?

The destination is similar in the main markets, the route differs. UK law has an unusual provision, section 9(3) of the Copyright, Designs and Patents Act 1988, that assigns authorship of a "computer-generated" work to the person who made the arrangements necessary to create it. On its face that looks like it could hand you a copyright in a pure AI image. In practice it sits awkwardly next to the modern requirement that a work be the "author's own intellectual creation," and in its March 2026 response on copyright and AI the UK Government proposed removing protection for works created solely by AI while keeping it for AI-assisted works (a proposal in a consultation response, not yet enacted law). The direction of travel is toward the same place the US already is.

The EU anchors copyright in the author's own intellectual creation, a standard built around human free and creative choices, so a purely machine-made image does not clear the bar there either. The throughline across all three: human creative input is the thing copyright attaches to, and a bare prompt is not it. (This is general information, not legal advice, and the law here is moving; check the current position for your jurisdiction.)

What this means for AI influencers and character work

If the single render is not a copyrightable asset, then the asset has to be something else. It is. For anyone running a virtual persona, the value was never one image you could have stopped others from copying anyway. The value is the persistent identity: the same recognizable character holding steady across hundreds of posts, the body of work, the audience that follows the persona, the name and the styling that make it that character and not a generic face.

That reframe is worth holding onto. A pure render is cheap and freely copyable by law. A consistent character, run as an ongoing presence, is hard to reproduce and is the part that actually compounds. Cladegrove is built around that persistence layer, holding one character stable across every render rather than treating each image as a one-off. If you want the mechanism, how Cladegrove locks a character's face covers the architecture, and the companion piece on how to disclose AI content covers the labeling side of running an openly synthetic persona. The labeling itself splits into the platform rules on Instagram and YouTube and, for an EU audience, the EU AI Act deepfake rules.

For the full picture on running disclosed, consistent AI characters, see how Cladegrove approaches AI disclosure.

Common questions