What is a personalised AI celebrity video?
It is one recording, licensed from a celebrity, used to generate many versions that each address a specific person. The celebrity records once; synthesis handles the variable portion, so a hundred thousand customers can each receive a video that uses their own name without the celebrity recording a hundred thousand times.
The nearest familiar comparison is a mail merge, except the output is video and the sender is someone the recipient recognises. That combination is why it performs: personalisation lifts response rates in every channel it has ever been applied to, and video is the format people actually finish.
Is this legal, and is it the same as a deepfake?
No, and the distinction is consent. A deepfake synthesises a person without their permission. Everything here begins with a written licence from the celebrity covering synthesis of their face and voice, for a stated use, on stated channels, for a stated term. Without that licence there is no project.
This matters more in India than it did a few years ago. Courts here have moved to protect public figures' personality and publicity rights specifically against unauthorised AI reproduction, and the direction is towards stronger protection rather than weaker. A brand that commissions synthetic media of a real person without documented permission is exposed regardless of who built it — so the licence is not paperwork, it is the thing that makes the campaign safe to run.
Our position is simple: we will not synthesise anyone who has not agreed to it in writing for that specific use, and we will not extend an old licence to a new use without going back to them.
What do you need from the brand?
A customer list with the fields you want personalised, a channel to deliver through, and a clear trigger. The list is the part to plan properly: personalisation fails loudly when the data is wrong, and a video that greets someone by the wrong name is worse than one that greets nobody.
You also need a lawful basis to message those customers, which is your existing CRM consent rather than anything specific to this format. Personalised video does not change who you are allowed to contact — it changes what you send them.
How many personalisation fields should you use?
Fewer than brands expect. One or two, placed naturally, read as a message meant for that person. Five, inserted mechanically, read as a database talking — and the uncanny effect undoes the warmth the celebrity was hired to supply.
Name plus one contextual field is the reliable combination: name and city, name and renewal date, name and product. The second field is what proves the message is specific rather than merely templated.
Should the video say it is AI-generated?
Our recommendation is yes, visibly, and the reasons are practical as much as ethical. Expectations around labelling synthetic media are tightening across platforms and regulators, and a label added now costs nothing while a campaign retrofitted later costs a rebuild.
It also tends not to hurt. Recipients understand that a celebrity did not personally record a video for each of them; the appeal is that the brand went to the trouble, not that anyone was fooled. Labelling removes the one outcome that would genuinely damage the brand — a customer discovering it themselves and concluding they were being tricked.
