How AI Avatar Videos Could Change Social Media

Synthetic presenters are the part of generated video with the least settled etiquette, and the gap between what is technically possible and what audiences accept is wider here than anywhere else. Worth separating the cases, because they get discussed as one thing and are not.

Three different things

A stylised avatar of you. An illustrated or animated version of a real person, obviously not photographic. Long-established — VTubers built an entire culture on it — and uncontroversial, because nothing is concealed.

A photorealistic synthetic person. Someone who does not exist, presented as a presenter. This is where almost all the friction is.

A generated likeness of a real person. Their face, saying something they did not say. Legally and ethically the most serious, and the category most likely to be regulated first.

Conflating these produces bad arguments. They have different risks and different answers.

What audiences actually object to

Not the technology. The discovery.

A talking head carries an implicit claim: a person is telling you this, and by appearing, vouching for it. When a viewer finds out nobody did, the reaction is not "interesting production technique" — it is the feeling of having been handled.

Which produces a reliable asymmetry:

• Told upfront: mild curiosity, occasionally interest in how it was made
• Discovered later: a sense of having been deceived, applied to everything else that account has said

The cost is not in the clip. It is in the retroactive doubt cast over the rest.

Where it is fine

• Obviously stylised presenters. Animated, illustrated, clearly not a photograph. No concealment, no problem.
• Your own avatar, disclosed. A consistent generated presence for a creator who is open about it.
• Explanatory content where nobody is vouching for anything. A generated narrator over an explainer is closer to a font choice than to an endorsement.
• Accessibility and localisation. Generated presentation in languages a creator does not speak, where the alternative is nothing.

Where it is not

• Anything implying personal experience. A synthetic face saying "I tried this" is a false claim, however good the clip.
• Testimonials and reviews. The whole value is that a person had the experience.
• Anything financial, medical or safety-related, where trust in the speaker is the mechanism.
• A real person's likeness without explicit consent. Regardless of how convincing.

The practical rule

Make it evident, not discoverable.

Disclosure buried in a caption technically satisfies a policy and does nothing for the reaction, because the discovery still happens — just later, and with the added insult of "it was disclosed".

Evident means visible in the clip: a stylised presenter, an on-screen label, a format that does not pretend. That costs nothing and removes the failure mode entirely.

The uncanny valley is not the problem any more

An assumption worth retiring, because it shapes bad predictions.

For years the reassuring thought was that synthetic people look slightly wrong, and that the wrongness is a natural warning system. That was true and it is becoming less so — not because the models perfectly solved faces, but because short vertical clips watched on a phone are a forgiving format. Four seconds at feed size hides most of what would be obvious at full screen for a minute.

Which means the safeguard people expected to be technical is not going to be technical. It will be provenance and norms: metadata attached at creation, platform labelling, and audience expectations about what a format implies.

That is a slower and less satisfying answer than "you will always be able to tell". It is also the realistic one, and it is why the practical advice here is about being evident rather than about how convincing the clip is.

What changes for everyone else

Two effects, both already visible.

Unsourced video is treated as unverified. Audiences increasingly default to scepticism about video from accounts they do not know. That is a rational adjustment, and it is not going to reverse.

Established trust becomes more valuable. When anyone can produce a convincing talking head, the scarce thing is having a reason to be believed — a track record, a real identity, a history of being right. That is the opposite of what people assumed cheap production would do.

The same dynamic applies to short-form content generally, and is worked through in how short-form AI video is reshaping TikTok and Reels.

If you are considering it

1. Decide which of the three things you are doing. They have different answers.
2. Choose evident over discoverable. Stylised if possible.
3. Never imply experience you did not have.
4. Check platform policy, which changes often.

The narrower question of where a stylised avatar belongs versus a photograph is in AI avatar vs professional photo, and the broader honesty framing in the ethics of AI-generated photos.

Kitana makes stills and short clips from your own photos rather than synthetic presenters — the avatar tool is among the thirteen in the apps, and Photo to Video animates a still you supply.