Account

What new tool do you want?

Text & Writing

AI Image Prompt Library & Builder

Build detailed prompts for Midjourney, DALL·E, Stable Diffusion and other image models — subject, style, lighting, camera and mood — with ready-made examples.

How to use

  1. 1Describe the subject, then pick a style, lighting, framing and mood.
  2. 2Copy the prompt into your image generator. Generate again for variations.

How it's calculated

Effective image prompts layer: subject → setting → style/medium → lighting → camera/framing → mood/color → quality details.

Frequently asked questions

Do different image models need different prompts?

Mostly the same ingredients work everywhere. Some tools add their own parameters (like --ar 16:9 in Midjourney) — add those at the end.

Should I name artists in prompts?

Describe the style (e.g. 'loose watercolor with ink outlines') instead of naming living artists — it's more respectful and often more precise.

Why add lighting and camera terms?

They strongly shape the look: 'golden hour, 85mm portrait lens, shallow depth of field' gives a very different result from 'flat studio light'.

A worked example of the layered structure

"a red fox sitting in a snowy forest" with Photorealistic style, Golden hour lighting and Close-up portrait framing assembles into: "a red fox sitting in a snowy forest, photorealistic photograph, highly detailed, natural textures, warm golden-hour sunlight, long soft shadows, close-up, 85mm lens, shallow depth of field, cohesive color palette, high detail." Each dropdown doesn't just insert a single word — it inserts a small cluster of related descriptive phrases (style brings 3 descriptive terms, lighting brings 2, framing brings 2-3) that image models respond to more reliably together than any single adjective would alone, since these models were trained on real photography and art captions that describe scenes exactly this multi-layered way.

Why the layering order mirrors how photographers and illustrators actually describe images

The fixed order — subject, then style/medium, then lighting, then camera/framing, then a closing mood/quality phrase — isn't arbitrary; it mirrors the convention used in real photography and art direction, where you establish what's being depicted before specifying how it's rendered. Image generation models pick up on this structure because it matches the pattern in the captions they were trained on, which is part of why swapping only the style dropdown from Photorealistic to Watercolor while keeping subject, lighting and framing identical produces a coherent result rather than a confused mix — the model recognizes the slot each phrase is filling.

Why naming artists is discouraged in favor of describing technique

The FAQ's guidance to describe style ("loose watercolor with ink outlines") rather than name a specific living artist reflects both an ethical concern — generating work explicitly mimicking a named artist's style without consent is a live controversy in AI art — and a practical one: describing the actual visual qualities you want (brushstroke texture, color palette, line weight) gives the model more precise, unambiguous information to act on than an artist's name does, since the model has to internally translate "in the style of X" into those same visual qualities anyway, often inconsistently.

Surprise me picks one named option, and a blank subject stops me

The subject opens on a red fox in a snowy forest. Style opens on photorealistic, lighting on golden hour, and framing on a close-up portrait. Aspect ratio opens on don't add, so nothing about a ratio is appended until I pick one. Choosing surprise me on style, lighting, or framing picks one of the named choices for that menu, so two runs can differ. A blank subject stops with "Describe the subject." A ratio I do pick is added at the end as an --ar flag. I leave the ratio on don't add when I want the prompt to end on the detail phrase.

Related tools

More: Text & Writing