How to Write Midjourney Prompts — A Beginner's Guide (2026)

If you’ve just opened Midjourney and typed “a cool dragon”, you already know the problem: the results are fine, but they’re nothing like the images you’ve seen other people share. The difference is rarely talent — it’s structure. Midjourney rewards prompts that are built in a particular order, and once you know that order, you can describe almost anything and get a usable image within a few attempts.

Two corpus numbers set expectations: the median tested Midjourney prompt is 52 words long, and 72% of our 71 Midjourney entries carry a --no exclusion chain (full corpus data).

This guide covers the structure we use for every Midjourney prompt in our library, the three parameters worth learning first, and the mistakes that waste the most time.

The anatomy of a Midjourney prompt

Nearly every strong prompt has the same five layers, in roughly this order:

1. Subject — what the image is of. Be concrete: “a lone lighthouse keeper” beats “a person”.
2. Setting & mood — where and when: “on a storm-battered cliff at dusk”.
3. Style — the medium or aesthetic: “cinematic photography”, “loose watercolor”, “flat vector illustration”.
4. Lighting & camera — the layer most beginners skip, and the one that makes images look intentional: “volumetric fog, backlit, wide shot”.
5. Parameters — the --ar, --stylize and --no switches at the end.

You don’t need every layer every time, but when a result feels generic, the fix is almost always adding the lighting/camera layer or tightening the subject.

Start from a working example

The fastest way to learn is to copy a prompt that already works, generate it, and then change one thing at a time. Here are two from our library that demonstrate opposite ends of the style spectrum:

Notice how the mood is carried by specifics — “rain-slicked asphalt”, “volumetric fog”, a named color palette — rather than adjectives like “beautiful” or “epic”, which do nothing.

This one shows style-first prompting: the medium (“loose watercolor”, “wet-on-wet wash”, “pigment blooms”) comes before the subject even matters. Swap the koi for any small subject and the painting treatment survives.

The three parameters that matter

Midjourney has many parameters, but three of them cover 95% of real use:

• --ar (aspect ratio). The default square is rarely what you want. Use --ar 16:9 for desktop wallpapers and thumbnails, --ar 2:3 for Pinterest and posters, --ar 9:16 for phone wallpapers and stories.
• --stylize (0–1000). How much creative license Midjourney takes. Low values (50–150) follow your words literally — best for logos and product shots. High values (300–600) produce more artistic, opinionated images — great for paintings and concept art. The default is 100.
• --no (negative prompt). A comma-separated list of things to exclude. For flat design work, --no photorealism, gradient, 3d, shading is the difference between a logo and an illustration — see it working in our minimal fox logo prompt.

Five mistakes that waste your generations

1. Adjective stacking. “Stunning, beautiful, epic, highly detailed masterpiece” adds nothing. One concrete visual detail beats five empty superlatives.
2. Contradicting yourself. “Minimalist” and “intricate details” in one prompt forces Midjourney to pick one — and you won’t control which.
3. Ignoring aspect ratio. Composition changes dramatically with the frame. A prompt tuned at 1:1 often falls apart at 16:9 until you re-describe the scene.
4. Fighting the model with re-rolls. If three generations in a row miss, the prompt is the problem. Change the wording instead of rolling again.
5. Forgetting to save what worked. When a prompt lands, save the exact text and settings. Future you will not remember.

How do I get consistent results?

Treat prompting like a recipe, not a slot machine. Change one variable per generation, keep notes on what each change did, and build a small personal library of phrasings that reliably work for you. That’s exactly how the prompts on this site are made — each one lists the exact settings and version we tested, because “works on my machine” applies to AI models too.

Where to go next

Browse the full Midjourney prompt library to find a starting point close to what you want to make, or explore by style — watercolor and cyberpunk are good first stops. Copy a prompt, swap the subject, and you’re already prompting better than most.