Who owns an AI-generated video? The UK and US answer, and how you prove your part
Updated 17 September 2026
What you own in a video made with AI is the part a human shaped, and it holds only as far as you can show that. A film you directed, cut and graded can be yours in the UK and the US even if AI helped. A clip generated by a machine and used unchanged is the weak point in both. The records of what you did and what you had permission to use turn ownership into something you can defend.
I have made films for brands and charities for fifteen years, mostly with real people in front of real lenses. I have only recently started using generated video, so what follows is not a lawyer's opinion and not a claim to have solved every case. It is the working process I use to keep three things apart on every job. What a person made, what a machine made, and what we had permission to use. One thing changed when generated shots started turning up in my edits. The question clients ask at the end of a job used to be "when can we have the files?" Now it is "do we actually own this?" In 2026 both the UK and the US moved on that question, and most of what you will find online was written before they did.
Picture a situation that is now ordinary. A charity commissions a short film. The interviews are real, shot in a community centre with a two-person crew. The edit needs the town at dusk and there was no time to get it, so the editor generates the shot. One interview has rough audio, so an AI tool cleans it. A stock track is licensed for the music. The film goes out and does well. Six months later a national partner wants to run it on television, and their lawyer sends three questions. What did the AI tools' terms allow? Who owns the film and the generated shot? Did you have the rights to everything in it?
Nobody did anything wrong. But three different rules now sit inside one two-minute film, and the answers depend on whether anyone kept the records. That is what this page is about.
Explore this guide
If there is one section to start with, begin with How you prove you made it. It is the part nobody else covers and the part that decides most disputes.
- Can you copyright a video made with AI? At a glance
- Three questions hiding inside who owns it
- Who owns AI-generated content in the UK
- Who owns AI-generated content in the US
- Who owns a commissioned video?
- The prompt is not the proof
- How you prove you made it
- What technology can prove, and what it cannot
- What to do when something goes wrong
- Questions people ask
- Where to go next
This page is a filmmaker's working understanding, not legal advice. The rules differ between countries and are still moving, so for a live dispute or a large campaign, get a lawyer. Primary sources are linked where each claim is made and listed at the end. This page does not cover EU labelling law, the lawsuits about training data, or the terms of individual tools. It links to the pages that do.
Can you copyright a video made with AI? At a glance
| Situation | UK | US |
|---|---|---|
| A film a person directed, cut and graded, with AI used along the way | Protected. The film is a work in its own right and its authors are the producer and principal director. | Protected. Human authorship is required and present. Assistive AI use does not weaken it. |
| A clip generated from a prompt and used as it came out | Uncertain. Section 9(3) may apply, but it is untested for generative video and the government has proposed removing it. | Not protected. Prompts alone do not make a human the author. |
| A generated clip you then reworked into a sequence | Your creative changes and the film around them can be protected. | Your selection, arrangement and modification can be protected where they reflect creative judgement. |
| A commissioned film | The maker is the first owner unless the contract assigns it. The client normally has a licence for the agreed purpose. | The client can own it from the start as a work made for hire, if the signed agreement says so. Otherwise the maker owns it unless assigned. |
| What the tool's terms give you | Permission from the vendor to use the output. Not copyright. | Permission from the vendor to use the output. Not copyright. |
| What proves your claim | Records of the human choices, the inputs and the permissions. | The same, plus disclosing generated material if you register. |
Three questions hiding inside "who owns it"
A contract with the vendor.
- What the platform lets you reuse
- Set by the service terms, which change
- Permission to use, not ownership
What most people mean by ownership.
- Rewards human creative choices
- Scope differs by country
- Follows what you actually shaped
Whether you were allowed to use it.
- Permission for what went in the frame
- Faces, voices, music, brands, locations
- The layer that most often triggers takedowns
When someone asks who owns an AI-generated video they are usually asking three questions at once, and each has a different answer.
The first is what the tool lets you do. Every generation tool has terms for its output. Those terms are a contract between you and the vendor. They can give you broad permission to use, sell and publish what you generate. What they cannot do is create copyright that the law does not recognise. A vendor can promise not to sue you. It cannot promise that anyone else will treat the output as yours.
The second is copyright, which is what most people mean by ownership. Copyright rewards human creative choices. In both the UK and the US, the parts of a video a person shaped can be protected, and the parts a machine produced on its own are, at best, in doubt. The next two sections set out where each country now stands.
The third is what went in. A generated shot can still contain a real person's face, a voice that sounds like someone, a brand's logo or a line from a song. Owning the output does not settle whether you had the right to use the inputs. That is a consent and licensing question, and it is the one that most often ends in a takedown. I cover faces and performances in how consent and likeness work when AI is used in film production and voices in what I see from the side that hires voice actors.
Keep those three apart and most of the confusion goes. You can have generous tool terms, a strong claim in your edit, and still be in trouble because of what was in the frame.
It helps to know which kind of AI video you have made, because the copyright answer gets stronger as the visible human work increases. AI used to clean up, sharpen, denoise or colour-correct real footage is assistive and changes nothing about ownership. A generated element placed inside a film you cut is a component, and the film around it is yours. A generated sequence you selected from many outputs, composited, retimed and graded carries your creative work in it. A clip generated from text and used unchanged is the one to worry about. My AI policy sorts the same material into assisted, altered and synthetic for disclosure purposes, and the two views line up.
Copyright protection tracks the visible human creative work. The more you shaped, the stronger the claim.
This shows relative strength, not a guarantee. Ownership is decided case by case, on the facts and the country.
Who owns AI-generated content in the UK
Start with the ordinary rule. Copyright in a creative work belongs to the person who made it, and it exists because a human made creative choices. That rule has not changed and it covers most of what a filmmaker does.
Now the point that matters most for video, and the one most articles miss. A film is a work in its own right. The Copyright, Designs and Patents Act 1988 defines a film as a recording on any medium from which a moving image may be produced, and makes its authors the producer and the principal director. It does not have to pass the originality test that applies to a drawing or a script. So the finished film you produced and directed is protected as a film, and the interviews you recorded, the edit, the grade and the arrangement carry their own protection inside it. A generated establishing shot on its own is a different and much weaker claim. Build your position on the film you directed and cut, not on the orphaned clip inside it.
That leaves the special case. Section 9(3) of the same Act says that where a literary, dramatic, musical or artistic work is generated by computer with no human author, the author is taken to be the person who made the arrangements necessary for its creation. It gives 50 years of protection and no moral rights, so it is a thinner right than ordinary authorship. It only applies to those four categories of work, not to films. It has almost never been tested in court, and lawyers have argued for years about how a work with no human author can be original at all.
On 18 March 2026 the government published its Report on Copyright and Artificial Intelligence, following the consultation that ran from December 2024 to February 2025. It proposes removing section 9(3) protection for wholly computer-generated works, noting that no other country has anything like it, and confirms that works made with AI assistance, where a person made the creative choices, stay protected as they always were. No law has changed yet. The proposal is on the table, not on the statute book, and the report says the government will keep monitoring the provision before acting.
So for a generated clip used as it came out, treat both the existence and the ownership of copyright as uncertain. Section 9(3) is still in force, it has not been tested on generative video, and it may go. For the film around that clip, the ordinary rules apply and your claim is as strong as your records.
One case gets cited more than it deserves. Getty Images v Stability AI, decided by the High Court in November 2025 and now under appeal, is not a case about who owns an AI-generated film. Getty lost its main copyright claim because a trained model does not store copies of the images it learned from, and won a narrow trade mark point about outputs carrying its watermark. Getty had dropped its claims about the training itself before the end of the trial, so the court never decided whether scraping images to train a model infringes copyright here. Treat it as a case about training and outputs that decided less than the headlines said.
Who owns AI-generated content in the US
The US answer is simpler and stricter. Copyright requires a human author. A work generated entirely by a machine cannot be registered and nobody owns the copyright in it.
That has been the Copyright Office's position for years, and the courts have backed it. In March 2025 the Court of Appeals for the D.C. Circuit held in Thaler v Perlmutter that the Copyright Act requires a human author, in a case where the applicant had listed his own AI system as the sole creator of an image. On 2 March 2026 the Supreme Court declined to hear his appeal. The Supreme Court did not rule on the question itself. It left the appeals court's decision standing, which makes human authorship the settled requirement unless Congress changes it.
The harder question is how much human input is enough. The Copyright Office answered that on 29 January 2025 in the second part of its report on copyright and artificial intelligence, and four of its conclusions matter to anyone making video.
Prompts alone do not make you the author. The Office treats a prompt as an instruction that conveys an idea, and ideas are not protected. A detailed prompt still does not control how the model turns it into pixels, and the same prompt can give a different result each time. A shot you generated and used as it came out is not yours to copyright, however carefully you described it.
Your own work stays yours when it is visible in the output. Feed in your footage, your drawing or your script, and if the result carries it, you keep copyright in your contribution.
Selection, arrangement and modification can count. Choosing among outputs, sequencing them, combining them with other material and changing them can be protected where those choices reflect creative judgement and add up to a sufficiently original edit. Picking one output from several uncontrolled results is not, on its own, enough. The Office decides this case by case. This is where an edit becomes an authored film, and where your records do the work.
Assistive use does not spoil anything. The Office's own example is a film company using AI for colour correction, sharpening or de-blurring in visual effects. That is assistive, and it does not weaken copyright in the finished film.
If you register, you disclose. Since March 2023 the Copyright Office has required applicants to disclose AI-generated material in a work and to exclude it from the claim where it is more than trivial, while describing the human contribution. You register the human-authored film. You disclaim the untouched generated parts. That is the operational answer to "can I copyright this", and it is another reason to keep the records this page describes.
Where the line falls between a heavily prompted image and an authored one is still being argued. Allen v Perlmutter, about an image refined through more than 600 prompts, was still awaiting a decision in a federal court in Colorado as of September 2026. Until it lands, the safe assumption is that prompting is not authorship, and editing can be.
Who owns a commissioned video?
In the UK, the maker owns a commissioned film unless the contract assigns it to the client. In the US, the client can own the film from the first frame if the signed agreement makes it a work made for hire. In both countries, a generated shot the maker does not own cannot be assigned to anyone. Now the detail.
For a commissioned film in the UK, the creator or production company is normally the first copyright owner. Paying for the film does not change that. The client will usually have a licence, express or implied, to use the film for the purpose it was commissioned for, but a licence is not ownership. If the client wants to own the film, the contract has to say so in an assignment. The one exception is a film made by an employee in the course of their job, where the employer is the first owner unless the contract says otherwise. That matters to agencies and production companies with employed editors and directors.
In the US, a commissioned audiovisual work is one of the categories the law allows to be a work made for hire. When it is, the commissioner owns the copyright from the start, not by transfer. Two conditions have to be met. The parties have to agree in writing, signed by both, that the work is made for hire, and the work has to fall within the listed categories, which a commissioned film does. If those conditions are not met, copyright starts with the maker and moves to the client only through a written assignment. That is the opposite of the UK default, and it catches people out in both directions.
The generated components sit inside that. A generated shot used without meaningful human change may attract no copyright in the US, and its position in the UK is uncertain. The surrounding film still carries the protected human work, the edit, the recorded interviews, the sound and the arrangement. So a contract that says "the client owns everything" transfers everything the maker has, and the maker cannot hand over a right they never had in the untouched clip. A careful client wants to know which parts were captured, which were generated and which were altered, and the contract should say what the client gets in each case.
The tool's terms are the third layer. Whoever generated the shot did so under a licence from the vendor. That licence may allow commercial use, may restrict it, and may have changed since the shot was made. A contract cannot override what the tool allowed. Read the terms on the day, keep a dated copy, and name the tools in the agreement.
If you are the one paying for the film, six questions settle most of this before it starts.
Does the contract assign the copyright to us, or give us a licence, and for what use?
Which parts of the film will be captured, generated or altered, and will we get a schedule that says so?
Which AI tools will be used, and will we get a dated copy of their terms?
Is any real person's face, voice or performance involved, and who holds the releases?
Who approves the disclosure wording, and where will it appear?
If a claim arrives after launch, who responds first, and what records will we have?
The full checklist is in how to commission AI-assisted video safely, and that page is where it lives. The one thing I now put in every agreement is a one-page schedule that lists what was captured, what was generated and what was altered. The template is further down this page.
The prompt is not the proof
Here is where the two halves of this page meet. In both countries, what you own depends on what a human did. In neither country does anyone check. There is no register you file with when you make a film and no inspector who watches you edit. If a question comes, from a client, a platform, a partner's lawyer or a court, the only thing that shows what you did is the record you kept.
People get this backwards in one particular way. They keep the prompts and throw away everything else, as if the prompt were the creative act. It is the least useful thing you can keep. In the US it is an unprotected idea. In the UK it is evidence of an instruction, not of authorship. What proves authorship is what shows selection, arrangement and change. Which shot you chose out of twelve. Where you put it. What you did to it afterwards. The prompt shows you asked. The edit shows you decided.
How you prove you made it
I call this the proof trail. It is one folder per project, kept from the first day, and it is what I would hand over if anyone asked me to show what was real, what was generated, what was changed and what I had permission to use. Each item does a different job. Together they answer the lawyer's three questions from the top of this page in minutes rather than weeks.
| What to keep | What it proves | What it looks like in practice |
|---|---|---|
| Camera originals | That the real footage is real | The untouched files off the card, with their metadata, kept apart from the edit |
| The project file and edit history | Selection, sequencing and structure. The core of authorship in both countries | The timeline saved at each milestone, with dated exports |
| Generation records | What was asked for, what came back, and what was chosen | Prompts, dates, the tool and version, the rejected outputs as well as the one used |
| Before and after | That you changed the generated material, and how | The raw output beside the finished shot. One still or a five-second clip |
| Grade, comp and audio work | Creative control over the final image and sound | Saved node trees, adjustment layers, key stills, the cleaned and uncleaned audio |
| The tool's terms on the day | What the licence allowed when the shot was made | A dated PDF or screenshot of the terms, not a link, because links change |
| Releases and approvals | Permission for people and brands in the frame | Signed releases, approval emails, the scope agreed |
| Licences for music and stock | That the inputs were cleared | The licence and the receipt, filed with the asset they cover |
| The contract and the schedule | Who owns what, and which parts were captured, generated or altered | The signed agreement plus the one-page schedule |
| Delivery and publish record | First use, approved version, what went out | The delivery date, the file hash or version name, the first link |
The minimum, if you keep nothing else. The project file, the camera originals, a short generation log, the before and after for any shot you generated or altered, and the releases and licences. That covers authorship, inputs and permissions, which are the three questions this page began with.
Two habits make the rest easy. Name generated and altered shots as such in the timeline, so the record builds itself as you cut. And write the one-page schedule at delivery, while you still remember, not when a lawyer asks. The disclosure lines I use, and the records I keep on every job, are set out in my AI policy, and this page does not repeat them.
The delivery schedule, and a generation log
This is the one-page schedule I attach to the delivery. Headings only. Copy it and fill it in.
| Section | What goes in it |
|---|---|
| Captured | Shots and audio recorded with a camera or microphone, with dates and locations |
| Generated | Shots, audio or elements produced by an AI tool, with the tool and version |
| Altered | Real material changed by AI, with what changed (extended, cleaned, replaced, de-aged) |
| Licensed | Music, stock and fonts, with licence reference and holder |
| Released | People and brands in the frame, with release reference and scope |
| Tool terms | The tools used and the date of the terms PDF on file |
| Disclosure | The disclosure wording used and where it appears |
And one line from a generation log, so you can see how little it needs to be.
| Field | Example |
|---|---|
| Asset | SHOT_014_GEN_TOWN_DUSK |
| Tool and version | Tool name, plan, model version |
| Date generated | 4 June 2026, 21.40 |
| Inputs | Text prompt. Reference still of the real high street |
| Outputs reviewed | 9 |
| Output used | Take 7 |
| Human changes | Trimmed to 3 seconds, retimed, graded to match adjacent shot, sky replaced |
| Rights checks | No people. No signage. Stock music separate |
| Disclosed | Yes. "Contains AI-generated imagery where indicated" in end card and description |
What technology can prove, and what it cannot
There is now a layer of technology built to travel with a file and say where it came from. It is useful and widely misunderstood, so it is worth being exact.
Content Credentials, built on the C2PA standard, attach a signed record to a file that says which device or software made it and what was done since. Some cameras can write these credentials at capture. Some editing tools carry them through an edit. Google's SynthID puts an invisible watermark into generated images and audio. In May 2026 OpenAI joined the C2PA steering committee and began adding SynthID to its generated images, and from July 2026 to its audio. Video coverage across the industry is still patchy. Since May 2026 YouTube has added its own label to videos it judges to contain significant photorealistic AI, whether or not the uploader disclosed it. The film industry's Human Provenance in Film standard, launched at Cannes in May 2026, is building a disclosure tier system on top of all this, with a consultation open until 31 October 2026.
Here is what that layer does for you. It can show that a real shot came out of a real camera. It can show that a generated shot was generated. It can back your records with a signature that is hard to fake. If your camera or your software can write credentials, turn them on.
Here is what it does not do. It does not stop anyone copying your work. A screenshot, a screen recording or a re-encode can strip credentials off a file, which is why the trust signals on a video tend to fail in a set order once it is reposted, as I explain in how audiences judge a video before they ask who made it. It does not tell you who owns the copyright. And it does not replace the proof trail, because a credential says a file was made in a certain tool at a certain time. It does not show the twelve shots you rejected or the contract that assigned the rights.
Provenance technology proves origin. Your records prove authorship. You want both.
What to do when something goes wrong
Most ownership trouble does not start in court. It starts as a platform flag, a client email or a message from someone who recognises themselves. A calm, quick reply with documents behind it settles most of it before it becomes anything else.
Capture the evidence first. Links, screenshots, timestamps, and the exact version being challenged.
Assemble the proof trail. Project file, originals, generation log, licences, releases, and the schedule.
Reply to the platform or the client with the documents and a short plain account of what you made, what was generated, and what changed.
If there is real commercial harm, a recognisable person, or a brand involved, get a lawyer in before you say more.
If the challenge is about a face, a voice or a performance rather than ownership, treat it as a consent question first. Disclosure does not cure a missing release, and speed matters more with real people than with anything else on this page.
Questions people ask about owning AI-generated video
Who owns AI-generated content in the UK?
The person who made the creative choices, the same as any other work. A film is protected as a film, with the producer and principal director as its authors. For material generated by a machine with no human author, section 9(3) of the 1988 Act currently names the person who made the arrangements as the author, but the government proposed removing that in March 2026 and it has never been tested on generative video. Treat purely generated material as uncertain and build your claim on what you did to it.
Who owns AI-generated content in the US?
Nobody, if it was generated entirely by AI. US copyright requires a human author, a rule the D.C. Circuit confirmed in 2025 and the Supreme Court declined to revisit on 2 March 2026. The human parts of a mixed work, including your selection, arrangement and modification of generated material, can be yours.
Can you copyright a video made with AI?
Yes, where a person shaped it. A film you directed, cut and graded can be protected in both countries even if AI helped, and the US Copyright Office says assistive uses like colour correction or de-blurring do not weaken that. What you cannot copyright is a generated clip used exactly as it came out of the tool.
Do I own an AI-generated video if I only wrote the prompts?
No, not in the US, where the Copyright Office treats a prompt as an unprotected idea, and it is uncertain in the UK, where section 9(3) is untested and proposed for removal. Your ownership starts where your editing starts.
Do I own a Runway, Sora or Veo clip I generated?
You have whatever permission the tool's terms give you, which usually covers using and publishing the clip. Whether you own copyright in it is a different question, and for an unchanged clip the answer is no in the US and uncertain in the UK. Keep a dated copy of the terms, because they change, and build your claim on what you did with the clip afterwards.
When does editing AI output become my own work?
There is no single step that flips the answer. Your claim gets stronger as your creative decisions become visible in the result. Substantial modification, compositing, timing, arrangement, and integration with real footage and sound all help. Picking one output from several is not enough on its own. The analysis is case by case, and a saved project file is the best evidence you can have.
If I commission a film in the UK, do I own it?
Not automatically. The maker is the first owner unless the contract assigns the copyright to you. Without an assignment you will normally have a licence to use the film for the purpose you paid for, which is not ownership. If it matters, put the assignment in the contract.
Who owns a commissioned video, the client or the production company?
Whoever the contract says. In the UK the maker owns it unless the contract assigns it, and the client otherwise has a licence for the agreed use. In the US it can be a work made for hire, owned by the client from the start, if the signed agreement says so and the work fits the listed categories, and otherwise it stays with the maker unless assigned. With no contract, expect an argument.
Can I register an AI-assisted film with the US Copyright Office?
Yes, for the human-authored parts. Since March 2023 applicants have to disclose AI-generated material and exclude it from the claim where it is more than trivial, and describe what the human author contributed. You register the film you shaped and disclaim the untouched generated clips. Your generation records make that form easy to fill in.
What rights do the AI tool's terms actually give me?
Permission from the vendor, and only that. The terms can let you use, sell and publish the output. They cannot give you a copyright the law does not recognise, and they can change after you have made the shot. Keep a dated copy.
Can I use AI-generated video commercially?
Usually, subject to the tool's terms and to what is in the frame. The bigger risks in commercial use are inputs rather than ownership, such as a face that resembles a real person, a voice, a logo or a protected character, and in the UK the advertising rules apply however an ad was made. That side is covered in my AI policy.
If someone re-uploads my AI-assisted video, what can I do?
Start with the human-authored parts you can document. Your edit, original footage, script, sound and graphics can support a claim, and a platform complaint based on them is stronger than one based on an untouched generated clip. Platform procedures vary and no result is guaranteed. A first-publish record and the project file are what make the difference.
What is the simplest proof trail I can keep without slowing production down?
One folder per job with the project file, the camera originals, a short generation log, one before and after for each generated or altered shot, and the releases and licences. Name generated shots as such in the timeline and write the one-page schedule at delivery. That is enough to answer any question on this page.
Where to go next
Own the decisions, keep the record
Making a polished video is getting easier every month. Showing that you made it, and that you were allowed to use everything in it, is getting more important at the same rate. The law in both countries now points the same way. Your ownership is your judgement, made visible. Keep the record of that judgement as you go and most of the questions on this page answer themselves.
What changed in this version
This page replaces the version published in February 2026. Since then, the UK government proposed removing section 9(3) in its report of 18 March 2026. The US Supreme Court declined to hear Thaler v Perlmutter on 2 March 2026, leaving the human-authorship rule in place. Getty v Stability AI was decided in November 2025 and is under appeal. YouTube began labelling photorealistic AI video automatically in May 2026, and the provenance tools described above moved from pilots to products. This page was reviewed on 17 September 2026 and will be updated when the Getty appeal is decided, when Allen v Perlmutter is decided, or if section 9(3) is legislated.
- 29 January 2025 The US Copyright Office publishes Part 2 of its AI report. Prompts alone are not authorship, but selection, arrangement and modification can be.
- 4 November 2025 Getty v Stability AI is decided in the High Court. The main copyright claim fails, the training question is left open, and the case goes to appeal.
- 2 March 2026 The US Supreme Court declines to hear Thaler v Perlmutter, leaving the human-authorship rule in place.
- 18 March 2026 The UK government proposes removing section 9(3), the provision that protects wholly computer-generated works.
- May 2026 YouTube begins labelling photorealistic AI video automatically, and the provenance tools move from pilots to products.
Primary sources
Copyright, Designs and Patents Act 1988, section 5B, definition of a film
Copyright, Designs and Patents Act 1988, section 9, authorship, including films and computer-generated works
Copyright, Designs and Patents Act 1988, section 11, first ownership, including employees
Report on Copyright and Artificial Intelligence, UK government, 18 March 2026
Copyright and Artificial Intelligence, Part 2, Copyrightability, US Copyright Office, 29 January 2025
US Copyright Office AI hub, including the March 2023 registration guidance
Thaler v Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025); certiorari denied 2 March 2026, No. 25-449
Getty Images v Stability AI [2025] EWHC 2863 (Ch), 4 November 2025; permission to appeal granted 16 December 2025
Human Provenance in Film, consultation open until 31 October 2026
