Describe who you need
Age range, wardrobe, expression, setting, and how tightly the shot is framed. "Woman in her fifties in a garden centre apron, mid-laugh, waist-up" beats "happy woman" every time.
Describe a person and get back a photoreal still of them. This is casting, not cloning — the model composes a plausible human from your description rather than reproducing anyone real, which is precisely what makes the output safe to run as paid creative.Describe a person and get back a photoreal still of them.
Age range, wardrobe, expression, setting, and how tightly the shot is framed. "Woman in her fifties in a garden centre apron, mid-laugh, waist-up" beats "happy woman" every time.
GPT Image 2 renders the person at the aspect ratio you pick. Run the prompt again with one detail changed to get a second casting option rather than a second pose.
Download the still, or send it into Creatives to add the offer, the logo and the placement crops the buy actually needs.
Use cases
Stock libraries are thin everywhere outside the 25-to-35 bracket, and thinner still once you need a specific occupation, region or body type. Describing the person you actually sell to is faster than filtering twelve thousand results for someone who nearly fits.
Most e-commerce brands have flawless packshots and no human imagery at all, which caps how far a prospecting set can be tested. A generated presenter gives the set a face to test against the pack shot without commissioning a lifestyle shoot.
The same layout, the same offer, a person who reads as local in each market. Re-running one prompt per region is the cheap version of the shoot you were never going to fund in six countries.