The whole thing, explained without the marketing

You upload a handful of photographs of your face from different angles. A model learns what is constant about that face. It produces new portraits of it in circumstances you did not photograph.

That is the mechanism, complete.

The consequence that decides your result

The system separates what is stable about you from what is circumstantial. If every source photograph came from one afternoon in one shirt, the shirt is stable — so it becomes part of you.

Different days, different clothes, different light. That single input decision matters more than which product you choose.

What prompts do and do not control

Style and setting: background, formality, framing. Not likeness. No phrasing rescues a bad input set.

Where it fails

Strong prescription glasses, tightly curled hair, darker skin tones, pronounced asymmetry — consistently, across vendors.

And the failure is subtle: the headshot looks real, it just looks like someone who is almost you. That gap, not realism, is the actual limitation.