How to Create AI Prompts for Design Inspiration

For a photograph you describe a moment. For a design you describe a set of rules: how wide the content column is, how many columns the grid has, what the spacing rhythm is, how many type sizes exist and how far apart they are, how much of the surface is empty. Rules are what let you produce a second and third piece that belong to the first.

A useful test: could someone rebuild your layout from the prompt without seeing the reference? If not, you have described a mood rather than a design.

Describe the page as an ordered stack of bands — header, hero, proof, feature sections, pricing, closing call to action, footer — and say what each band contains and how it is arranged. Then describe the system: content width, section padding, corner radius, border and shadow behaviour, button shape and states, card style, and the palette with concrete values including which colour is background, which is surface, and which is the single accent.

The clause that matters most is the accent restriction. Interfaces read as designed when colour is rationed and applied consistently, and models default to sprinkling it everywhere unless told not to.

Start with format and margin, then hierarchy. Say which element is largest and by roughly how much, where the eye should land second, and what the typographic contrast is — condensed against a wide serif, all caps against a lowercase italic, one family at three weights. Describe the image treatment separately from the type: duotone, halftone, full bleed, boxed with a hairline rule.

Prompt for a direction rather than a logo. Describe the personality in concrete terms — "industrial, precise, slightly severe" — and translate it into the mechanics: geometric versus humanist letterforms, tight versus open tracking, whether the palette is neutral with one accent or fully chromatic, whether shapes are hard-cornered or rounded, whether photography is documentary or staged.

Do not ask for another company's identity. Describe what you liked about it instead; the description is what you can legitimately reuse and build on.

Social work is constrained by crop and legibility. Say the aspect ratio, state the safe margin, cap the word count, and specify the minimum type size relative to the frame — for example "headline occupies at least a fifth of the height".

Interface work is constrained by state. Describe the density, the control shapes, the empty and loading states, and how focus and disabled states look. A prompt that only describes the happy path produces a screenshot, not a design.

Weight (light to black), width (condensed, normal, extended), case, tracking, leading, the size ratio between levels, and family contrast. For composition: grid columns, gutters, margins, alignment, rhythm, density, the focal point and the reading path. These words are boring and precise, which is exactly why they work.

If you already have a page or poster you admire, the website and poster tools will write this vocabulary out for you from the reference — palette with hex values, type scale, spacing and section order — and you can edit from there. See also how to write better prompts and the design examples.