What a dedicated generator does that a general model cannot

Almost everyone tries the same thing first: a general chat model, one photograph, and a request for a professional headshot.

Why it disappoints

It produces a plausible portrait of a person. It does not produce a consistent likeness of you, because from one photograph there is nothing to learn about what is constant in your face across angles, expressions and light.

The difference is architectural rather than a matter of quality. A dedicated system trains on a set of your photographs specifically to separate what is stable about you from what is circumstantial.

What that implies for the input

Source photographs from different days, in different clothes, in different light. Fifteen from one afternoon in one shirt teach it that the shirt is part of your face.

No amount of prompt wording rescues a bad input set — prompts control background and formality, not likeness.