Real people vs synthetic avatars in AI ads: what converts, and what's legal

By HyperKnown Team·Published September 1, 2026

AI video gives brands two ways to put a human face in an ad without a shoot. One is the synthetic avatar: a stock or generated character from a library, available instantly. The other is a licensed real person: an identity-verified human whose likeness you license for the campaign, with the content generated from their verified likeness. They look superficially similar in a feed. They differ on almost everything underneath: rights, platform treatment, disclosure, audience response, and what happens when something goes wrong.

What's the difference between a synthetic avatar and a licensed real person?

What converts better?

The honest answer is that no universal number settles it, and anyone quoting one without context is selling something. What performance teams consistently observe is a mechanism, not a constant: social ad formats reward content that reads as a person talking, and penalize content that reads as an ad. That is the entire logic of UGC-style creative, and it is where the difference between the two approaches shows up.

A synthetic presenter is generic by design: the same faces appear across thousands of brands' ads, and feed audiences have learned to recognize the category. A licensed real person is specific: a face that exists once, attached to a real identity an audience can find. Specificity is what authenticity-driven formats reward, and it compounds: a creator whose likeness performs for one brand builds a track record others can license. The takeaway is narrower than real always wins: the two approaches are different creative inputs, so test them as separate treatments rather than assuming parity.

What's legal, and what's merely permitted?

Both approaches can be run lawfully. The difference is where the risk sits. With a synthetic avatar, your rights chain runs through the library's terms and its upstream acquisition from whatever real people trained the model, none of which you can inspect. Restrictions on paid advertising or claim categories often sit in that fine print. With a licensed real person, the chain is one link long: the person approved your brief, and the record exists. The consent workflow piece details what that record contains.

Disclosure applies either way. Under the EU AI Act's transparency rules, and under major platforms' synthetic-media policies, AI-generated realistic humans must be labeled whether the underlying face is synthetic or licensed. Licensing a real person removes the consent question, not the disclosure one.

When should you use which?

• Synthetic avatars fit internal content, training material, high-volume low-stakes localization, and placements where a generic presenter is fine and speed is everything.
• Licensed real people fit paid social and UGC-style ads, anything where the presenter embodies the brand, campaigns making claims that will face scrutiny, and any placement where a journalist, regulator, or customer might ask who that person is and whether they agreed.
• Neither fits ads implying a real person endorses something they never saw. With a synthetic look-alike of an actual person, publicity-rights liability applies as if you had used the person directly.

How does licensing a real person actually work?

On HyperKnown, brands browse identity-verified creators, submit a brief with the campaign scope, and the creator approves or declines it. Payment happens only after approval, the license tier defines the cleared channels from social through CTV, and every project leaves a consent record. The creator receives 65% of the license fee, which is why real people keep joining the supply side. The complete guide covers the model end to end.