Midjourney prompts look different from prompts for other image models. Where Nano Banana and GPT Image read full sentences, a Midjourney prompt is a line of short visual phrases followed by parameters such as --ar 3:2 --stylize 250. Get the order and the parameters right and you spend your time choosing between good images instead of rerolling bad ones.

This guide covers the structure we use, the parameters worth learning, and examples for common jobs. The image prompt generator writes prompts in this format when you pick Midjourney as the target.

The shape of a Midjourney prompt

A Midjourney prompt has two parts: the description, then the parameters.

Midjourney
an elderly watchmaker at a cluttered workbench, small attic workshop, close-up over the shoulder, warm desk lamp light against cool window light, documentary photography, quiet focused mood, brass gears and loupe in sharp focus --ar 3:2 --stylize 200

The description is a run of comma-separated phrases. We order them like this:

  1. Subject: who or what, with one or two defining details.
  2. Setting: where it is.
  3. Composition: framing and camera angle (close-up, wide shot, low angle, top-down).
  4. Lighting: source, direction and colour.
  5. Style or medium: photo, watercolour, isometric 3D, ink drawing, film still.
  6. Mood: one or two words.
  7. Detail: the small thing that makes it specific.

Short phrases beat long sentences here, and concrete nouns beat adjectives. "Warm desk lamp light against cool window light" tells Midjourney more than "beautiful dramatic lighting". Leave out filler such as 8k, masterpiece or ultra detailed; it adds nothing you can control.

The parameters that matter

Parameters go at the very end, each starting with two hyphens. Nothing descriptive should come after them.

--ar: aspect ratio

Sets the shape of the image as width:height. The default is square.

  • --ar 1:1 for profile pictures and square posts
  • --ar 4:5 for portrait posts
  • --ar 3:2 for classic photo framing
  • --ar 16:9 for wallpapers, thumbnails and cinematic frames
  • --ar 9:16 for phone wallpapers and Stories

Ratios must be whole numbers, so write 37:20 rather than 1.85:1.

--stylize (--s): how much Midjourney's own taste applies

Ranges from 0 to 1000, with 100 as the default. Low values follow your words more literally; high values add Midjourney's own sense of composition, colour and polish, sometimes at the expense of what you asked for. Our generator writes values between 100 and 400, which keeps the prompt in charge while still looking finished.

--chaos (--c): variety between the four images

Ranges from 0 to 100, default 0. At 0 the four images in a grid are close variations. Raise it to 20 or 40 when you are exploring and want four genuinely different takes.

--no: things to leave out

A short negative list: --no text, people. Use it for one or two specific problems, such as unwanted lettering or a crowd you did not ask for. A long --no list works less well than describing what you do want.

--style raw: less house style

Turns down Midjourney's default beautifying. Useful for photography that should look like a real photo rather than a polished illustration of one.

--seed: repeatable starting noise

Any whole number. The same prompt with the same seed and settings gives similar results, which helps when you want to test the effect of changing one word.

--weird (--w): unusual results

Ranges from 0 to 3000, default 0. Small amounts, around 50 to 250, push the image toward stranger, less conventional choices. Large values get very odd very fast.

--v and --niji: model version

--v followed by a number picks the Midjourney model version. --niji picks the anime and illustration model instead. If you leave both out, Midjourney uses whatever version is set as default in your settings.

--sref: style reference

Points at an image URL (or a style code) whose look you want to borrow. Midjourney takes the palette, texture and overall feel without copying the content. --sw controls how strongly the reference applies.

Image prompts

To use a picture as part of the prompt, put its URL at the very start, before the text. The image steers composition and content, and --iw sets how much it counts against your words. If you only want the style of a picture, use --sref instead.

If you have the picture but not the words, image to prompt writes a Midjourney prompt from it, with --ar already matched to the picture's shape.

Examples for common jobs

Product shot

Midjourney
a frosted glass perfume bottle with a brass cap, standing on wet black slate, tight three-quarter close-up, single softbox from above and a thin rim light, studio product photography, luxurious and calm, water droplets on the glass --ar 4:5 --stylize 150 --style raw

Isometric illustration

Midjourney
a tiny ramen shop on a street corner at night, isometric view, warm lantern light spilling onto wet pavement, clean 3D illustration with soft shadows, cosy mood, steam rising from the bowls on the counter --ar 1:1 --stylize 300

Book cover background

Midjourney
a lone lighthouse on a cliff under a sky full of stars, wide shot with the horizon low in the frame, cold moonlight and a warm beam, painterly digital illustration, lonely and hopeful, empty sky at the top for a title --ar 2:3 --stylize 250 --no text

Exploring a direction

Midjourney
a greenhouse café filled with hanging plants, wide interior shot, soft morning light through fogged glass, editorial interior photography, peaceful, mismatched vintage chairs --ar 16:9 --stylize 200 --chaos 30

Common mistakes

  • Parameters in the middle. Anything after a parameter can be misread. Keep all parameters at the end.
  • A space or punctuation inside a parameter. Write --ar 16:9, not --ar 16 : 9 or --ar16:9.
  • Sentences and instructions. "Please make sure the woman is smiling" is written for a chat model. "smiling woman" does the job.
  • Fighting the stylize value. If results keep drifting from your description, lower --stylize or add --style raw before rewriting the prompt.
  • Pasting the same prompt into other tools. Other image models ignore or misread --ar and --stylize. Rewrite the prompt for the model you use.

Writing Midjourney prompts faster

The image prompt generator takes a one-line idea and returns a Midjourney prompt in the order above, with --ar set from the aspect ratio you choose and a sensible --stylize value. Switch the target to Nano Banana, GPT Image or FLUX and the same idea is rewritten as full sentences for that model. For finished prompts with our own renders, browse the image prompt library or the whole prompt library.