Real People in AI Video: Likeness Rights and What Platforms Allow

There is a moment in every AI video project where someone asks the question: can we just put the founder in it? Or the mayor, or a historical figure, or that actor whose face would make the whole thing land. The honest answer is more interesting than yes or no, and it decided the fate of one of our own festival entries.

We had an episode built around a historical figure. Strong concept, strong craft. It was not eligible. The contest rules blocked depictions of real people without authorization, and historical figures counted as real people. Nobody on the jury cared how respectful the portrayal was. The rule existed before our film did. That day taught us to read platform policy the way we read a script: before production, not after.

Likeness rights AI: what the law actually protects

Likeness rights, sometimes called right of publicity, protect a person's face, voice, name and recognizable identity from commercial use without consent. This is not new law invented for AI. It has protected celebrities from unauthorized endorsements for decades. AI just made violations cheap and fast.

Deepfake laws are now layering on top. Jurisdictions across the US, the EU AI Act, and legislation in Brazil and Asia are converging on the same principles: synthetic depictions of real people require consent, disclosure requirements are expanding, and penalties are moving from civil to criminal in cases involving elections, fraud or intimate imagery.

The practical rule for a production studio is simpler than the legal map suggests. If a real, identifiable person appears in your AI video and they did not sign for it, you have a problem. The size of the problem varies by country. Its existence does not.

AI video consent: image and voice are not the same problem

Here is a distinction we learned in production, not in a legal memo: image and voice carry truth differently.

A generated image of a person reads as an interpretation. Audiences have a century of paintings, caricatures and stylized portraits teaching them that a picture can be about someone without being them. A cloned voice reads as testimony. When a voice says words, the listener's brain files it as something that person said. That is why voice cloning triggers a sharper ethical and legal response than face generation, and why platforms police it harder.

So consent needs to be specific. Consent to appear is not consent to be voiced. Consent to be voiced reading a script is not consent to have new sentences generated forever. When we work with a client's founder, the release covers image, voice, the exact use, and the duration. Anything vaguer is a lawsuit on a timer.

If the depiction needs a disclaimer to be defensible, it usually needs a rewrite instead.

And one more thing we hold as studio doctrine: a disclaimer does not solve the ethics. Writing AI generated under a video of a person saying things they never said does not undo the saying. The audience feels the person speaking before they read the fine print.

What platforms and contests actually allow

Policies differ, but the pattern across major AI video platforms, festivals and contests is consistent:

• Living public figures: blocked or heavily restricted without documented authorization.
• Historical figures: increasingly treated as real people. Do not assume death equals public domain for a face.
• Private individuals: consent required, full stop, and platforms will remove content on complaint.
• Fictional characters performed by real actors: risky territory, because the actor's likeness rides inside the character.
• Your own team and clients who signed releases: the safe, and honestly more creative, lane.

Contests are stricter than platforms. A platform moderates after upload. A contest disqualifies before judging, and there is no appeal that beats the rulebook. Verify the rules at the source, on the day you commit to the concept, because these policies are rewritten constantly.

The direction decision: design people who do not exist

The constraint sounds limiting until you direct inside it. Our answer is to build original characters with the same rigor a casting director brings to real ones. A character sheet, a consistent face carried by reference across shots, a designed voice built legitimately rather than cloned from someone real. The audience connects to a person on screen. They do not need that person to have a birth certificate.

This is also where the work stops being a legal question and becomes a craft question. A synthetic spokesperson designed for the brand can be more precise than any borrowed celebrity, because every feature is a decision, not an accident of casting availability.

The checklist before anyone generates a frame

• Is any identifiable real person depicted, in face or in voice?
• Do we hold a signed release covering this exact use, medium and duration?
• Does the destination platform or contest allow it, verified at the source this week?
• If the answer to any of the above is shaky, can an original character carry the film better?

Question four is the one that matters. In our experience, the answer is almost always yes.

Consent for a face is not consent for a voice, a disclaimer does not launder either, and an original designed character almost always carries the film better than a borrowed real one.