Who is in frame: face rights in advertising made with AI

Who is in frame: face rights in advertising made with AI

The question "whose face is that" arrives a month after delivery, when the clip is already running. Below are three different questions people usually merge into one, three legal sources of an image, and four contract lines that close almost everything.

The question "whose face is that" arrives a month after delivery, when the clip is already running. Below are three different questions people usually merge into one, three legal sources of an image, and four contract lines that close almost everything.

---

The question that surfaces after delivery

You made an advertising clip with a character. The client is pleased and the clip is running.

A month later the question arrives: whose face is that?

And it turns out you had not thought about it. Took it from the generation, it looks good, it works. Where the model got it from is unclear, but it looks like a person.

If the person exists, that is somebody else's image. The complaint arrives not to you but to the client, and comes back to you with interest.

Let us take it apart properly: where the line is, what is allowed and what goes in the contract.

Three questions, not one

The subject usually gets reduced to "can AI faces be used". In fact there are three questions and they are different.

First: whose face is in frame. Does that person exist and did they consent.

Second: whose rights in the result. Who owns the generated material — you, the client, or nobody.

Third: what the client was told. Do they understand the person in frame is not a real customer.

The first is the most dangerous, the third the most underrated. In order.

Whose face: what is not allowed

Celebrities and public figures. In no form, with no caveats, not "in the style of". The right in an image belongs to the person, and being well known does not cancel it.

Photographs from the internet. What you found in a search belongs to the person in it and to whoever took it.

Stock photographs as a basis for generation. Even with a stock licence, it usually permits using the shot rather than training on it or reworking it into a new face. Licences have to be read, and in most cases the answer is no.

A face resembling a real person. The trickiest case. If a generated character is recognisably like an existing person, that is their problem with you rather than your good luck.

Whose face: what is allowed

Three legal sources. There are practically no others.

Your own face

A selfie, and you are the character in the advert. That works for an expert, a coach, an author — advertising with yourself in frame and no shoot.

The limit: it is your face. Selling the clip to a client, you are putting your own image into their advertising — and that has to be agreed: where, for how long, in what context.

Your own trained character

A permanent character trained by you and stored in your profile. The appearance does not change from frame to frame.

Why that is the best option for commercial work. The character does not exist, so nobody's rights get infringed. And it is your asset: it can be used in series, sold as part of a service or handed to a client with the project.

What matters here: do not train on somebody else's photographs. Otherwise you are back at the first prohibition.

Ready avatars with rights

The section's gallery holds ten characters for different roles: cosy and lifestyle, youthful, business and expertise, kitchen and home, style and grooming, minimalism and tech, maker and DIY, city and sport, experience and trust.

Every avatar was generated by us. They are not stock people and not somebody else's faces — no rights questions arise.

For a freelancer delivering work to a client that is the fastest safe route: you hand over the clip and do not think about whose face is in it.

A special case: a client's staff

It comes up constantly and gets handled badly if not thought through in advance.

The client brings photographs of their employees and asks for characters to be made from them for advertising.

What is needed: written consent from each person to the use of their image, including AI reworking and commercial use.

What usually happens: the client says "oh, they don't mind" and that is that. Six months later an employee leaves on bad terms and their face in somebody's advertising becomes a subject for discussion.

How to do it properly: the client obtains the consent, and you record in the contract that they confirm holding rights in the material supplied. That is one line, and it moves the responsibility to where it belongs.

The third question: what the client was told

The underrated point.

Staging is allowed. A false claim is not.

A virtual character showing a product is a normal advertising format. A virtual character saying "I bought it and it helped me" is a false review, because nobody bought anything.

The line is simple: the format can be staged, a fact cannot. You cannot attribute to a non-existent person an experience that never happened.

That is particularly strict in subjects with advertising requirements: medicine, finance, children's products. There, a review from a non-existent person is not a stylistic question.

What to do: agree with the client in advance how the text gets phrased. Not "I lost ten kilos" but a demonstration of the product and its properties.

What to put in the contract

Four points, one line each. They cover almost everything.

The source of the images. The characters are generated and do not reproduce real people, or a face the parties hold rights in is being used.

Responsibility for material supplied. The client confirms they hold rights in everything they gave you — product photographs, staff, logos.

Rights in the result pass on full payment. Not on presentation, not on handing over the files — after payment.

Labelling and advertising requirements. Who applies the synthetic content label and who is responsible for meeting advertising rules. Usually the client, but it has to be written down.

Why that is not bureaucracy. A contract without those lines means that on any complaint the freelancer ends up carrying it — simply because there is nobody else.

What to do if the work is already delivered

A practical case: the clips are running and the face came from who knows where.

Check. Search for similar images. If the character is recognisably like an existing person, that is a risk.

Change the source. A trained character or an avatar with rights. Clips get reassembled faster than a complaint gets resolved.

Warn the client. Silence here is worse than an awkward conversation: if the problem surfaces on its own, you are at fault twice.

What is available for this

Ten ready avatars with rights in Marketing Studio — all generated by us.

Character training in Gen AI — a permanent character stored in your profile and dropped into any generation.

The legal category — thirteen assistants, including an intellectual property specialist and a corporate lawyer. Going through a contract, checking wording, preparing consents. Chat access comes with the Basic plan.

A caveat specific to this subject: nothing legally significant goes out without review by a human lawyer. An assistant speeds up the analysis and prepares drafts; a person carries responsibility for the wording.

Registration is free and opens three days of full Basic access.

Where to start today

One action. Open your most recent piece of work with a person in frame and answer one question: where did that face come from.

If the answer is "it generated itself", check it for resemblance. If it is "found it online" or "took it from stock", that is the one mistake of all possible ones costing you not a rebuild but a legal complaint.

It changes in an evening: train your own character or take an avatar with rights. Everything else in this article is about never having to return to the question again.

---

What to do when the character has to be permanent

Rights are the first question. The second: where to get a face you can use forever.

The three legal sources are listed above. Practically the third is the most convenient — your own trained character, and in the ready solutions it is implemented like this:

Creating the character — you set a name, a biography and upload a reference photo that becomes the benchmark. From there the system holds facial consistency across every generation.

What that removes legally. The character does not exist, so nobody's rights get infringed. And it is your asset: used in series, handed to a client with a project, or kept as the basis of a service.

And a limit to keep in mind: you may train on your own material. A reference from the internet takes you back to the first prohibition.

---

What to read next

[How an AI influencer earns](/en/blog/ai-blogger-monetization) — where these rights are needed.

[AI accounts: what gets banned](/en/blog/ai-account-what-gets-banned) — the other five risks.

[AI for lawyers](/en/blog/ai-for-lawyer) — how to check a contract yourself.

[A legal risk audit](/en/blog/legal-risk-audit) — the same question at company level.

[AI video generation in 2026](/en/blog/ai-video-generation-guide) — the full guide to video models.