Type a person. Get a performer.

Prompt

Why this step

The brief is the cheapest part of the production and the bit most teams under invest in. A sentence written with care does the work of a casting agency, a stylist and a lighting director combined. The trick is not to describe a face. It is to describe a feeling someone has when they look at that face, and then let the model find the geometry that produces it.

Prompt

Why this step

An avatar is not an image. It is a small set of reference frames that the model can return to whenever you ask for this person again. Without this step you get a series of strangers who happen to look similar. With it, you get a performer with a face the algorithm and the audience both learn to recognise, which is the actual moat.

Prompt

Why this step

Variation without identity drift is the whole game. People remember the person before they remember the product, so the person needs to survive being moved from a sunny kitchen to a moody studio without quietly turning into a different human. This step is the rehearsal that prevents the audience noticing the seams.

Prompt

Why this step

Performance is mostly the micro expressions between the words. A locked face is the fastest way to fail the so called uncanny test, because viewers forgive imperfect skin long before they forgive a smile that does not move the eyes. A small library of credible expressions is what lets the same avatar carry a hook, a product reveal and a soft close in the same edit.

Prompt

Why this step

Most AI ads fail at the scene step because the model has not been told where on earth the human is. Naming the room, the light and the camera does more than naming the camera brand ever will. Once the avatar sits inside a believable place, the audience stops auditing the pixels and starts auditing the offer, which is exactly where you want them.