Who Owns an AI-Generated Image? Copyright, Explained
This question gets confident answers on the internet that do not survive contact with the actual law. The honest position is that ownership of AI-generated images is unsettled, differs substantially between countries, and is still moving. This article separates what is established from what is contested. It is general information, not legal advice — for anything commercially significant, ask a lawyer in your jurisdiction.
Two different questions
People conflate two things that are legally distinct.
The first is **copyright**: does anyone own an exclusive right to this image, and if so, who? The second is **usage rights**: what does the service you used permit you to do with the output? These have different answers, and in practice the second one governs most day-to-day decisions.
What is reasonably settled
Purely machine-generated images generally do not attract copyright in the United States. The US Copyright Office has consistently held that copyright requires human authorship. An image produced by entering a prompt and accepting the result has been refused registration on that basis. Where a human has made substantial creative contributions — significant arrangement, selection and modification — the human-authored elements can be protectable, while the machine-generated parts are not.
The UK has a specific provision for computer-generated works. UK copyright law contains a clause assigning authorship of a computer-generated work to the person who made the arrangements necessary for its creation. It predates modern AI by decades, its application here has not been comprehensively tested, and it has been the subject of consultation on whether to retain it.
The EU and India both require human authorship as a general principle, with the details of how that applies to generated output still being worked through.
Your own photograph remains yours. The photo you upload is a separate work with its own copyright. Nothing about generating from it transfers the rights in the original.
A person's likeness is protected separately from copyright. Personality rights, publicity rights and data protection law all apply regardless of who owns the image. This is the rule that actually constrains most real-world use.
What is contested
Whether training image models on copyrighted works constitutes infringement is being litigated in several jurisdictions, with decisions going different ways on different facts. Whether generated output can infringe a specific training image is a separate question, usually turning on substantial similarity to a particular work. And whether a detailed, iterated prompting process constitutes sufficient human authorship is precisely the boundary the cases are being fought over.
Nobody can tell you today how these resolve.
What matters in practice: the service terms
For most people, most of the time, the operative document is not copyright law — it is the terms of the service that generated the image. Those terms set out what you may do with the output, and they are enforceable as a contract whether or not anyone holds copyright.
UPretty's position, set out in its Terms, is that you retain ownership of images generated from your own photographs and may use them for personal projects, social media, professional profiles and most commercial purposes. Read the Terms page for the full statement, and read the equivalent page on any other service before relying on its output commercially.
The rules that apply regardless
These hold no matter how the copyright question resolves.
1. **Do not generate identifiable images of other people without their consent.** Likeness rights are independent of copyright, and this is where real legal exposure sits for ordinary users.
2. **Do not present generated images as documentary evidence.** A generated image of an event that did not happen is a fabrication, and framing matters more than intent.
3. **Do not imply endorsement.** Generating yourself in a setting that suggests a brand, employer or institution endorses you is a misrepresentation regardless of image ownership.
4. **Disclose where disclosure is expected.** Journalism, academic work, competitions and an increasing number of platforms require it. Several jurisdictions are moving toward mandatory labelling of synthetic media.
5. **Do not assume you can register it.** If your business model depends on holding an exclusive right in an image, generated output is a weak foundation in most jurisdictions today.
Practical guidance by use
Personal use — profile pictures, gifts, social posts. Effectively unconstrained, provided the person depicted has consented. This covers the overwhelming majority of use.
Commercial use where the image is incidental — a blog header, a background, a slide. Generally fine under most services' terms. Check the terms, and keep a record of which service produced the image.
Commercial use where the image is the product — stock licensing, merchandise, book covers you intend to enforce rights over. Get advice. The inability to hold copyright may matter a great deal here.
Anything involving another identifiable person. Get written consent, and get it before you generate rather than after.
The summary
In most countries, a purely machine-generated image probably belongs to nobody in the copyright sense. What you have instead is a contractual permission from the service to use it, plus whatever rights attach to your original photograph. For personal and ordinary commercial use that is entirely sufficient. For anything where exclusivity is the point, it is not — and that gap is unlikely to close soon.