Can you copyright AI-generated images and video?

This page covers United States law only, and it is a plain overview, not legal advice. The law is moving and differs by country, so check local law or ask a lawyer before you rely on any of it for a decision that matters. I have dated the facts below so you can see how fresh they are.

The biggest source of confusion is that three different questions get mixed together. Keep them apart and most of the confusion goes away.

Three different questions: copyrightable, usable and infringing

1. Is it copyrightable? Can anyone hold a copyright in the output, so that they can stop others copying it? This is a question of law, and it is what this page is about.
2. Can you use it commercially? Does the AI tool's licence or terms let you use the output in paid work? This is a question of contract between you and the tool maker. A tool can let you use an image commercially even if nobody holds a copyright in it. See can you use AI video in paid ads for clients.
3. Could it infringe someone else? Does the output copy someone's work, trademark, face or voice? This is about the rights of third parties, and it is a risk even when the first two answers look fine.

A tool saying "you own the output" does not settle question 1. A copyright registration does not settle question 3. Treat them as three separate checks.

The short answer (US): not on its own, but your human contribution can be

If you type a prompt and the AI produces an image or video, the US Copyright Office does not treat that output, by itself, as something you wrote. If you then add real human authorship, that part can be protected. The practical line is between what the machine produced and what a person created, chose or arranged.

What the Copyright Office said, and what the Supreme Court left in place

The Copyright Office. The Office published Part 2 of its report on copyright and artificial intelligence, on copyrightability, on 29 January 2025. Its press release says outputs can be protected only where a human author has contributed enough expressive elements. It says a perceptible human contribution, such as creative arrangement or modification of AI output, can qualify; that prompts alone do not; and that using AI as a tool to assist a human work, or including AI material in a human work, does not stop that work being protected. It did not recommend new law. Earlier, in a notice published in the Federal Register on 16 March 2023, the Office set out how people applying to register works must treat AI-generated material.

If it matters to your case, read the full report, a long PDF, at copyright.gov/ai. The description here draws on the Office's press release and its AI hub page, plus summaries by a copyright trade group and a writers' group, so details such as how the report treats repeated prompting may be incomplete.

The courts. Stephen Thaler, a computer scientist, tried to register an image made by his AI system with the AI named as the author and no human contribution claimed. The D.C. Circuit, a federal appeals court, held that copyright needs human authorship. On 2 March 2026 the US Supreme Court denied the petition to hear the appeal in Thaler v. Perlmutter (docket 25-449), according to the SCOTUSblog case page. A refusal to hear a case is not a new ruling, it leaves the appeals court decision in place. The question was narrow, because nobody claimed any human input, and so it does not tell you where the line sits for a person who prompts, selects and edits. As far as I could tell, there is no fixed percentage, and Congress had not acted.

What counts as human authorship in images and video

This is where business owners can act. Going by the Office's press release and the summaries, these are the kinds of human work that point toward protection:

• Your script, story and words. Writing is yours even when the picture is not.
• Your own photos, drawings or footage that you feed in as the base, when the final work keeps your expression.
• Your edits. Painting over, compositing, retouching by hand, cutting a sequence, adding motion graphics.
• Selection and arrangement. Choosing which pieces to use and how to order and layer them can be original. The Office has registered works that combined AI images with human text and arrangement, as I understand its decision on the comic book "Zarya of the Dawn". In that kind of case the protection covers the human parts and the arrangement, not each AI image on its own.
• Your voice and performance, if you recorded them.

What does not seem to be enough: a prompt, even a long and clever one, and picking your favourite of many results. The Copyright Alliance, a trade group, summarises the Office's view as treating repeated prompting like rolling dice again. Treat that as a summary, not a quote.

Why this matters for logos, mascots and ads

If a logo, mascot or hero video is pure prompt output, a competitor may be able to copy that exact output without breaking copyright, because there may be nothing to enforce. That does not mean they can pass themselves off as you: trademark and unfair competition rules are separate, and how they apply to your case is a question for a lawyer. Ask a lawyer about trademark for your name and logo.

What to do with the risk:

• For things you need to be distinctly yours, start from something human: a hand-drawn sketch, a photo you took, a character you designed in your own words and drawings.
• Use AI for the parts that matter less, such as backgrounds, b-roll and variants.
• Add your own layer: your script, your voice, your edit, your layout.
• For a mascot you plan to use for years, read who owns a branded AI character first.

What a contract can and cannot do

A contract can move whatever rights exist from one party to another, and it can say who may use the work and how. It cannot create copyright that the law does not recognise. So if a client contract says "you will own the copyright in everything we deliver", and part of the delivery is pure AI output, that sentence may promise more than anyone can give.

A safer set of promises, which a lawyer should confirm for your case:

• Assign whatever rights the agency or business holds in the deliverable, without saying that rights exist in every part.
• Give a licence under the AI tool's terms for the parts the tool made.
• List which parts were made with AI, so the client knows where protection may be thin. See should agencies disclose AI use.
• Limit warranties to what you know. Do not warrant that no one else will produce a similar image.
• Say what the tool does with uploaded files, since some tools may train on them. See does your AI tool train on client work.

The tool's terms are a separate layer. For example, Higgsfield (a web platform that gathers several AI image and video models in one place) has help-centre wording that its terms of use do not tie commercial use rights to a particular plan, while its pricing table marks commercial use as not included for the free plan and free output carries a watermark. Wording like this answers question 2 (use), not question 1 (copyright). Read the current terms of any tool you use, and for the free tier check what free AI video generators really give you. Never assume that an output is unique, or that no one else can have produced something close to it.

Practical steps: records, human layers, registration and wording

1. Keep records. Save the prompts, the reference images you supplied, the project files with your edits, and dated versions. The Authors Guild, a writers' group, advises authors to keep drafts and version history; the same reasoning applies to images and video. Records help you show what a person did.
2. Add visible human work where you want protection, and keep the layers separate so you can show them.
3. If you register, tell the Copyright Office about AI-generated material and disclaim it, and describe your human authorship. The 2023 registration guidance covers this. Do not claim the whole work if a machine made part of it.
4. Check for third-party risk. Look at the output for anything that resembles a known character, logo, brand, building or person. Do not use real faces or voices without written permission. See the consent checklist.
5. Be careful with words. Do not call AI output "original," "unique" or "fully protected" in a contract, a pitch or an ad.

Other countries: check local law

This page is about the US. Other places may treat AI output differently. The UK government ran a consultation on copyright and AI from 17 December 2024 to 25 February 2025. Its page says copyright protection for computer-generated works is among the issues, and that a report on the use of works in AI development was published on 19 March 2026. The page does not state an outcome on computer-generated works, so no outcome is guessed here. Official texts for the EU, China, India and other countries are not covered. If you sell or publish outside the US, check the rules there.

When to talk to a lawyer

• You plan to register a work that includes AI material.
• The output is central to your brand: a logo, a mascot, a signature video.
• You are signing or writing a client contract that promises ownership.
• Someone says your output copies their work, or you suspect someone copied yours.
• A real person's face, voice or likeness is involved.

How I'd approach this

I would treat AI output as raw material, not as a finished asset I own outright. For anything that has to be distinctly mine, I would start from my own sketch, photo or script, add my own edit, keep the files and dates, and avoid words like "unique" in anything I sign. For ads and social posts, I would not worry about copyright as much as about infringement, labels and consent. And before I promised a client ownership, I would ask a lawyer to check the wording.

Blur Studio builds the record-keeping side of this for small businesses and agencies: a shared asset library that logs the tool, prompt, reference files and approver for each piece. See the business page if that would help.