Prompt basics
How to Write AI Image Prompts: Beginner's Guide
Learn to write AI image prompts that work: a simple five-part formula, before-and-after examples and tips for Midjourney, Flux and Stable Diffusion.
By Let's Prompts · · 3 min read
An AI image prompt is the text you give a generator such as Midjourney, Flux, Stable Diffusion or GPT Image to describe the picture you want. The model never sees the idea in your head — only your words — so the prompt decides how close the result gets. The good news is that prompt writing is a learnable skill with a simple structure.
What a prompt actually does
An image generator turns your text into numbers and uses them to steer an image out of random noise. Every word nudges the result. Vague words leave decisions to the model; specific words make those decisions for you. That is why "a woman" returns something generic, while a description with a place, a light source and a lens returns something you can actually use.
The five-part prompt formula
Most strong prompts answer five questions, roughly in this order:
- Subject — who or what is in the picture, and what are they doing?
- Setting — where and when does it happen?
- Style or medium — photograph, 3D render, watercolor, anime, vector logo?
- Lighting and colour — golden hour, soft studio light, neon, muted pastels?
- Camera and composition — close-up, wide shot, 85mm lens, top-down, shallow depth of field?
You do not need all five every time. A logo needs style and colour more than a lens; a portrait needs light and camera more than a setting. Treat the formula as a checklist, not a template.
From weak to strong: a worked example
Here is one idea written three times. Notice what each version adds.
a woman in a city
a woman in a red coat walking through a rainy Tokyo street at night
Photorealistic portrait of a woman in a red wool coat walking through a rainy Tokyo side street at night, neon signs reflecting on wet asphalt, soft rim light, shot on an 85mm lens at f/1.8, shallow depth of field, cinematic colour grade, vertical 3:4 framing
Version 1 leaves everything to chance. Version 2 fixes the subject and setting. Version 3 also decides the medium, light, lens, colour and framing — which is why it is the one you can reuse and tweak. If portraits are your thing, the photorealistic portrait prompts in our library show this level of detail across hundreds of looks.
Use concrete words, not empty adjectives
Words like "beautiful", "amazing" and "high quality" tell the model very little. Swap them for things it can picture: "freckled skin with visible pores" instead of "realistic", "brushed aluminium with soft reflections" instead of "premium", "overcast morning light" instead of "nice lighting".
Change one thing at a time
When a result is close but not right, edit a single element and generate again. If you rewrite the subject, light and style together, you cannot tell which change helped. Keep notes on versions that worked — a personal prompt collection saves hours.
Match the prompt to the tool
Different generators read prompts differently. Midjourney responds well to short, evocative descriptions plus parameters like --ar 3:4. Flux and GPT Image handle longer, natural sentences. Stable Diffusion often rewards tag-style phrasing and supports negative prompts. Our comparison of Midjourney vs Flux vs Stable Diffusion prompts shows how to adapt one idea to each.
Learn faster from prompts that already work
The quickest way to improve is to read good prompts next to the images they produce. Browse the Let's Prompts library, copy a free prompt, then change one detail at a time to see what it controls.
Frequently asked questions
How long should an AI image prompt be?
Long enough to cover subject, setting, style and light — usually one to three sentences, or roughly 30 to 80 words. Past that point extra words tend to add noise rather than control.
Do I need to learn prompt engineering to use AI image generators?
No, but a basic structure helps a lot. Start with the five-part formula, then refine by changing one element at a time.
