GPT Image-2 UI mockup prompts

Designers were the early heavy users of GPT Image-2 for one reason: mockup generation. A good mockup prompt cuts the time between "we should explore an idea" and "we have something on the screen to argue about" from hours to seconds. The cases on this page are written exactly for that handoff.

A well-formed UI mockup prompt declares three things up front: the device frame (desktop browser at 1440 × 900, iPhone 15 Pro portrait, iPad landscape), the design system (Material 3, Apple HIG with rounded corners, custom flat with neutral grays), and the structural anatomy of the screen (left nav, top bar, content grid, modal overlay). The prompt then walks through each region with concrete content — not "a list of recent items" but "a list of three recent items: A, B, C, with timestamps and avatars."

The cases here go beyond shallow screenshot generation. We include dashboards with realistic data densities, dark-mode variants of the same screen, mobile-tablet pairs to test responsive thinking, design-system landing pages (the kind agencies pitch with), and edge-case screens like empty states, error states, and onboarding flows. These are the screens that actually matter in a real product, and they tend to be missing from generic prompt libraries.

A note on text inside mockups. The model handles Lorem-ish UI strings well when they are short ("Save changes", "Last updated 2h ago", "Add member"). Long body copy still wanders. If the screen requires real long-form text (a marketing landing page, a doc viewer), quote the exact paragraph in the prompt — or call out that the model should render plausible filler in a specific tone.

When you find a mockup you like, copy the prompt and paste it into your favorite generator. If you want to use the result in a real comp, open the prompt in a tool that exposes higher-resolution output (most hosted GPT Image-2 clients support 2x export).