How to write an AI art prompt: a beginner's guide with examples

A good AI art prompt names the kind of artwork first, then the subject, the style, the setting and the light, in one or two plain sentences. After that you generate, compare the results and change one thing at a time. The only real twist is that each tool reads prompts a little differently: Midjourney likes short phrases with settings tacked on the end, while Ideogram, ChatGPT and Gemini prefer full sentences.

By the end of this guide you'll have a reusable prompt formula and one finished prompt, in about 15 minutes. I built it from the official prompting guides for Midjourney, Ideogram, Google's Gemini image models and OpenAI's GPT Image models, checked in late October 2026. Nobody at the site ran hands-on tests for this piece, so treat every example as a starting point (honestly, that's how prompts work anyway).

Things you need before you start

• One image generator. Any of the four above will do; if you haven't picked one yet, our roundup of the best AI image generators compares free and paid options.
• A rough idea of where the picture's going: a poster, a phone wallpaper, a birthday card, a book cover.
• A notes file for prompts that work. Editing a good prompt is far quicker than starting again.

Step 1: Learn your tool's prompt format

Spend two minutes on this first. It saves a lot of confused results.

Midjourney's own guide says "Short and simple prompts typically generate the best images with Midjourney", and it warns that long lists of instructions can confuse it. Its parameters go after your text, with a space before the two dashes and no commas inside them. The current default model is V8.2, which became the default on 24 July 2026.

Ideogram is the opposite. It doesn't read --ar or ::1 weights as commands, so you set aspect ratio in the app. OpenAI suggests a steady order: scene, then subject, then key details, then constraints. For Firefly, an Adobe community expert in the Firefly forum says to "Aim for one or two sentences", ordered subject, environment, style.

Step 2: Write a one-sentence image summary

Start with the type of artwork and the main subject. Nothing else yet.

A storybook illustration of a red fox reading a book.

Ideogram's structure guide calls this the image summary: one sentence that describes the whole picture as if someone glanced at it for two seconds. If you only write one sentence, make it this one. Be exact with numbers, too; Midjourney points out that "cats" is vague and suggests "three cats" instead.

You should now have a line someone could sketch from. If they'd have to ask "what kind of picture?", you've skipped the artwork type.

Step 3: Pick an art style and a medium

This is what turns a generic picture into AI art you'd actually hang up. Name the medium (watercolour, gouache, linocut, pixel art, 3D clay) and then describe how it looks. Gemini's guide gives a useful pattern for stylised work: name the style, the subject and the visual qualities, such as bold outlines, cel-shading and the colour palette.

Midjourney's --niji parameter switches to a model focused on anime and Eastern aesthetics. For cartoon styles, I'd start with the sticker look or a 1930s rubber-hose cartoon (bendy limbs, pie-cut eyes), since both are easy to describe in plain words.

Watch out: I'd skip living artists' names entirely. Describe the brushwork, palette and era you like instead. It's fairer to working artists, and in my view it gives you more control, because you're naming the qualities you want rather than hoping the model copies the right ones.

A gouache storybook illustration of a red fox reading a book, with flat opaque colours, soft edges and a warm, limited palette.

Step 4: Add the setting, light and mood

Now say where the scene happens and how it's lit. Midjourney's checklist covers environment, lighting, colour, mood and composition; any of those will change the feel of the image more than another adjective about the subject will.

A gouache storybook illustration of a red fox reading a book by lamplight at the top of a lighthouse at night. Flat opaque colours, soft edges and a warm, limited palette of amber, deep blue and cream. The mood is calm and cosy.

Tip: use precise words. Ideogram's vocabulary list notes that "worm's-eye view" or "cranberry red" gets closer than "from below" or "red".

Step 5: Set the framing and the tool settings

Decide the framing (close-up, full body, wide view, bird's-eye), then set the shape: a parameter in Midjourney, a setting or plain request elsewhere.

Midjourney also has a stylize setting that runs from 0 to 1000, with a default of 100. Low values follow your prompt closely. High values give the model more artistic freedom, though it may drift from your details.

Gouache storybook illustration of a red fox reading a book by lamplight in a lighthouse at night, flat opaque colours, amber and deep blue palette, calm cosy mood, full-body view --ar 2:3 --s 150

You should now see a portrait-shaped image. If it's square, check there's a space before the dashes.

Step 6: Generate, compare and change one thing

Run the prompt and look at all the results side by side. Then change a single thing. OpenAI's guide recommends starting with a clean base prompt and making small, single-change follow-ups such as "make lighting warmer", and Gemini's advice is close: follow up with requests like "Keep everything the same, but change the character's expression to be more serious."

In ChatGPT and Gemini, just reply in the chat; elsewhere, edit and rerun.

Worked example: a storybook cover from start to finish

Say you want a cover for a children's bedtime story called "The Night Keeper". Here's the whole process, start to finish.

Start with the summary (a storybook illustration of a red fox in a lighthouse), then add style, scene and framing, keeping the fox low in the frame so the title has room at the top. The cover needs lettering, so I'd use Ideogram or ChatGPT here. Ideogram tells you to put the exact words in quotation marks, mention them early and keep them short; OpenAI says to put literal text in quotes or ALL CAPS and describe the typography.

A gouache storybook book cover. At the top, the title "THE NIGHT KEEPER" in large cream hand-lettered capitals. Below it, a red fox in a knitted scarf sits reading a book by lamplight at the top of a lighthouse, with a calm sea and stars visible through the round window. Flat opaque colours, soft edges and a warm palette of amber, deep blue and cream. The fox sits in the lower third of the frame. Calm, cosy bedtime mood.

Set a portrait ratio, then generate. If the title's misspelled, shorten it or spell the word out letter by letter, as OpenAI suggests for tricky words. Fox too realistic? Add "simple shapes, soft fur" to the style sentence.

Common mistakes and how to fix them

Saying what you don't want

Write "a party with no cake" and you may well get cake. Midjourney says so in its prompt basics, and Ideogram explains that models latch onto the word that follows "no" or "without". Describe the scene positively ("an empty street") instead. In Midjourney, use the --no parameter at the end; Ideogram 4.5 and 4.0 have no separate negative prompt field at all.

Contradicting yourself

"A minimalist sculpture with fine and intricate details" forces the model to pick one, as Ideogram's troubleshooting page points out. Choose one direction.

Burying the important detail

Words near the start of the prompt tend to carry more weight in Ideogram. If a detail keeps getting ignored, move it forward or mention it twice.

Writing a wall of text for Midjourney

Cut the "show me a picture of" and "please make it" parts. Keep the nouns and style words.

Tiny faces and hands

When a person's small in the frame, there are too few pixels for fine detail. Ideogram suggests framing words like "close-up" or "portrait" to move in.

Tips to get better results

• Say what the image is for. Google's guide says "Create a logo for a high-end, minimalist skincare brand" will yield better results than "Create a logo."
• Use Ideogram's Magic Prompt to expand a short idea, then turn it off if it keeps adding things you didn't ask for.
• If you're stuck for ideas, an AI prompt generator can give you a first draft to edit.

One small thing surprised me while reading these guides: OpenAI's advice actually encourages explicit exclusions such as "no watermark" or "no extra text", while Google and Ideogram push you toward positive phrasing. I think the safest habit is positive wording first, and a short exclusion list only if something keeps creeping in.

Next steps

The same method carries over to photos; our guide to writing an AI photo prompt goes deeper on lighting and camera language. To edit or upscale what you make, see our list of AI picture tools.

FAQ

How to write AI image generation prompts?

Start with the type of image and the main subject in one sentence. Then add the style, the setting, the lighting and the framing, keeping it to one to three sentences for most tools. Generate a few versions, compare them, and change one detail at a time until the image matches what you pictured.

How to write AI prompts for images?

Write what someone would see. "Orange and pink light on calm water" gives the model more to draw than "a beautiful scene". Put the main idea first and describe what you want rather than what you don't.

How to prompt AI art?

Name the art medium early: watercolour, linocut, pixel art, oil painting. Then describe the look in plain words, such as bold outlines, a muted palette or visible brush strokes. I'd avoid living artists' names. Describing the qualities you like is fairer to working artists and, I think, gives you more predictable results anyway.

How to write an AI art prompt?

Use this order: artwork type, subject, action, setting, lighting, style, framing. Midjourney wants short phrases plus parameters like --ar 2:3 at the end, while Ideogram, ChatGPT and Gemini handle full sentences well and can usually render short quoted text inside the image.

How to make good AI art prompts?

Be specific. Exact numbers, precise colour names and a named art style all help, and you should avoid contradictions like "minimalist but highly detailed". Save prompts that work and reuse their style sentences. Most good results probably come from three or four rounds of small edits, not from one perfect first attempt.