GPT Image-2 poster prompts

Poster prompts are the hardest category to write well, and the most useful when they work. The reason: GPT Image-2 is improving fast on typography, but a sloppy poster prompt still produces flat layouts with floating, half-legible text. A tight poster prompt produces something you can hand to a print shop.

The cases in this collection are written with a strict layout grammar. Each prompt names the format (A2 / A3 / theatrical one-sheet), the type hierarchy (display headline, sub-headline, ribbon, footer credits), and the typography family in concrete language ("condensed grotesque sans, weight 700 for the headline, italic serif for the credits"). It also names the color logic — usually one dominant + one accent + one neutral — so the result doesn't drift toward the mid-tone mush that generative models like to default to.

You will see four sub-genres here. Editorial covers borrow from magazine systems: hierarchy first, then imagery. Event posters are deliberately loud, with a single hero element and a strong slab below it. Movie keyart leans cinematic, with the title locked to the bottom third and characters at high contrast. And type-driven posters skip imagery entirely — they treat the headline itself as the visual.

Every prompt quotes the headline text explicitly. If you swap the quoted text for your own copy, the typography usually holds. If you remove the quotes and describe the headline in vague terms, the model invents words. Keep the quotes; change the content inside them.

The Chinese-language poster cases are written with CJK typography in mind — the model handles 楷体, 宋体, 黑体, and 隶书 reasonably well when you name them and pick a tight composition. Mixed-script posters (Chinese headline + English subhead) work, but ask for them explicitly or the model will pick one and drop the other.