Start with a small visual brief
A useful structure is outfit, expression, then mood. For example: “cream jacket over a black shirt, subtle silver chain, relaxed smile, warm amber light.” Each phrase describes something that can appear on screen.
Keep the adult protagonist’s identity tied to the uploaded portrait. Asking for a celebrity, a different person or a new face conflicts with the purpose of the photo reference.
Direct movement with restraint
Try “gentle shoulder movement, confident glance to camera, natural hand gestures.” A small set of movements can fit a continuous take more easily than a long choreographed sequence.
Avoid simultaneous opposing camera instructions such as locked-off framing and a rapid orbit. Do not ask to edit, extend or replace the reference footage; the workflow generates a new fictional scene.
Revise one detail at a time
After a first render, decide whether the outfit, expression or framing needs the most improvement. Keep the other directions steady and change one phrase, so the next result is easier to evaluate.
The prompt limit is 1,500 characters, not a target you need to fill. Leave out links, private information, unsafe content and requests involving people without permission. A short, precise prompt can be enough.