Aster Lane: a skincare set
GPT Image 2 · 4 pictures · 4:5 and 1:1
A serum bottle that has to look the same in a hero shot, a macro, a face-care frame and a flat-lay. The hero was made first, and the other three were made from it as the anchor.
Words that did the work: The one thing that must not change (bottle shape, colour, label) sits in a closing sentence of its own, and every frame has a named ratio. Swap the surfaces and the palette and the same structure works for any single product.
What we saw: The bottle's shape, cap and label carry through the hero and the flat-lay. The frosted glass the prompt asked for looks more like matte green plastic.
See the replay
AURELIA: a watch listing
GPT Image 2 · 4 pictures · 4:5
One watch in four frames, each with its own words on the picture. The prompt says to make the hero first and wait, then make the other three from it.
Words that did the work: It says outright that scene and text may change and the product may not. That sentence, plus a chained reference, is how a listing keeps one product across several shots.
What we saw: Every quoted string is right: the wordmark and tagline, the three spec labels (40mm case, Swiss movement, 5ATM water resistant) and the macro caption. The watch keeps its case, dial and strap; the hands sit at different times. No extra text appeared.
See the replay
A jacket on three people
GPT Image 2.5 · 4 pictures · 1:1 flat-lay, then 3:4
A flat-lay of a camel chore jacket, then three invented people wearing it. Each follow-up opens with the same guard: keep the exact jacket from the reference, and do not redesign it.
Words that did the work: The list of garment features that must survive comes before the description of the person, so the model reads what is fixed first.
What we saw: Colour, corduroy collar and three patch pockets hold. The button count does not: the prompt said four and the flat-lay has five. If a number matters, say it twice and count afterwards.
See the replay
Splash Lab: food and drink ads
GPT Image 2 · 4 pictures · 1:1
Four products, each described in the same six labelled fields: environment, hero, materials, motion, light and output.
Words that did the work: Because every prompt has the same fields, a difference in the result traces back to the one field you changed. The "8K" line works as a style cue: the pictures are 1254 pixels square.
See the replay