DALL·E / GPT Image

Verified against ChatGPT · 2026-08-10

Design a multi-screen app UI concept that keeps its own design system straight across screens

A phased UI-concept brief that locks a design system in the first generated screen — corner radius, type scale, accent color — and carries it into follow-up screens as in-conversation continuations instead of restating the whole brief and risking a different design system on every screen.

ChatGPT (GPT Image)5 fillable variables

The prompt

Ready to copy — highlighted parts are example details you can swap.

PHASE 1 — ESTABLISH THE SCREEN AND THE DESIGN SYSTEM

APP CONCEPT
A budgeting app for freelancers that tracks irregular income against fixed monthly bills

SCREEN TO DESIGN FIRST
the home dashboard, showing this month's income so far against upcoming bills

DESIGN SYSTEM TO ESTABLISH
rounded 20px corners on cards, a single teal accent (#1F8A70) used only for primary actions and the income number, generous whitespace, a clean geometric sans for numbers and a softer sans for labels. State these as fixed rules for every screen this conversation goes on to produce, not a style suggestion for this one image — corner radius, type scale, spacing rhythm, and the specific accent color should read as deliberate, repeatable system choices, not a one-off illustration style that happened to look good once.

DEVICE FRAME
a modern iPhone frame, portrait orientation, shown as a realistic device mockup rather than a flat rectangle

Generate this first screen now as a clean, high-fidelity concept render — real-looking UI chrome (status bar, navigation, buttons) with legible label text rendered directly in the image, not lorem-ipsum blocks standing in for text.

PHASE 2 — ADD SCREENS THAT REUSE THE SAME SYSTEM (same conversation, after Phase 1 is approved)
the bill-list screen and the add-income screen. For each one, explicitly instruct: reuse the exact type scale, corner radius, spacing, and accent color from the approved first screen — reference "the same design system as the screen above" rather than restating the whole design-system description again, so the follow-up generation anchors to what was already approved instead of independently reinterpreting a generic app style for each new screen.

PHASE 3 — CATCH DRIFT BEFORE IT COMPOUNDS
After each new screen comes back, check it against the first screen specifically for: the accent color hasn't shifted hue, the corner radius on cards and buttons matches, the type weight and hierarchy match. If any screen drifts, correct it as an in-context edit ("match the button corner radius to the first screen exactly") rather than accepting the drift and hoping the next screen self-corrects — small inconsistencies compound across a multi-screen set faster than they get noticed one screen at a time.

WHAT NOT TO DO
Do not ask for all screens in one single combined image — a single generation asked to render several different app screens at once reliably shrinks each one down toward illegibility and blends their UI chrome together. Request one screen per generation, in sequence, in this same conversation.

OUTPUT
One high-fidelity concept image per screen, delivered in sequence, each one visibly sharing the same type, color, and shape system as the first.

Customize

Optional — swap in your own details for the highlighted parts above.

Why this works

Within one ChatGPT conversation, a follow-up image request can condition on the actual pixels of the previously generated screen rather than only on language describing it, which is the mechanical reason anchoring Phase 2's screens to "the same design system as the screen above" holds a specific accent hex and a specific corner radius steady in a way that re-typing "teal, rounded corners" into an independent new prompt cannot — a fresh generation has no rendered reference to match against and can only approximate the same words again, which is exactly where a second screen's teal drifts a few degrees off the first screen's teal without anyone asking for that. Explicitly naming the design-system tokens up front matters because GPT Image, like any general image model, defaults to whatever generic mobile-app visual style is statistically common in its training data unless told otherwise, and that default look is not guaranteed to repeat identically across two separate generations even for the same stated app concept. The instruction against combining several screens into one image targets a documented compositional limit: a single frame asked to fit multiple distinct UI screens divides its available detail and legible-text budget across all of them, so each individual screen's labels and controls render smaller and less crisply than the same screen would generated on its own — sequencing one screen per generation keeps every screen's UI text at a size the model can actually render cleanly.

What you get back

A high-fidelity iPhone-framed dashboard screen with rounded teal-accented cards and legible labels, followed by a bill-list screen and an add-income screen generated afterward in the same thread, each one carrying the identical corner radius, teal accent, and type pairing established in the first screen.

Verified against

ChatGPT GPT Image (2026 release) · 2026-08-10

Changelog

  • 2026-08-10 Initial publish, verified against ChatGPT GPT Image (2026 release).

Need this built into your business?

If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.

EXPLORE WHAT SCULT BUILDS
All DALL·E / GPT Image prompts

Check your AI visibility

One URL in, a 0–100 score and the exact fixes out.

RUN THE CHECK

Browse all the tools

15 tools across six categories
13 of them never send your data anywhere

Free · No signup · No trial clock

SEE THE DIRECTORY