GPT Image 2 prompt templates

Twenty-two industrial prompt templates for GPT Image 2, each one a fill-in-the-blank form rather than a prompt to copy verbatim. They were reverse-engineered from a public catalog of more than 500 shared GPT Image 2 results — posts where the creator published the image and the prompt together — by looking at what the successful prompts in each genre had in common and turning that into a set of decisions you fill in.

The difference matters. A prompt you copy verbatim produces one picture, and adapting it means guessing which words were load-bearing. A template makes that explicit: the brackets are the decisions — subject, layout, palette, aspect ratio, the exact strings that must appear in the image — and everything around them is the structure that held the original result together. Fill the brackets, and you get a result in the same genre with your own content in it.

The templates are grouped the way the underlying catalog is grouped: interfaces, charts and infographics, posters and typography, products and e-commerce, brand systems, architecture, photography, illustration, characters, narrative scenes, historical themes, document layouts, and a catch-all for the experimental cases. Most categories carry more than one block — a standard version, one or more specialised variants, and a JSON version written for agents, where the same decisions appear as key–value fields that a script or an assistant can assemble and iterate on field by field.

Every template page carries the same sections: when to use it, the template itself with a copy button, how to fill it in, the mistakes that break it, real example results with links back to the creators who posted them, and the sibling blocks from the same category in case a neighbouring variant fits your brief better. None of it promises a particular output — model version, reference images and settings all move the result — but it does remove the guesswork about what a prompt in that genre needs to say.