Midjourney Parameters Guide

The full parameter table

Add these to the very end of your prompt, separated by spaces. Order does not matter. Where a parameter has a short form (like --s for --stylize), both work the same.

--ar : aspect ratio

Sets the shape of your image as width:height. The default is 1:1, a square. Use --ar 16:9 for widescreen scenes, --ar 9:16 for phone screens and stories, and --ar 2:3 for posters and portraits. Midjourney decides the shape here; the final pixel size comes from upscaling. Ratios wider than 2:1 still work but are treated as experimental.

--v : version

Picks which Midjourney model draws your image. Newer versions read prompts better and render more detail. As of 2026 the range runs 1 through 7, and --v 7 is the default. Swap in --niji instead of --v to use the Niji model, which is tuned for anime and illustration, for example --niji 6.

--stylize / --s

Controls how hard Midjourney pushes its own polished look. Low numbers (like --s 0 to 50) stick closely to exactly what you typed. High numbers, up to 1000, make images look richer and more artistic but drift further from your words. The default is 100.

--chaos / --c

Decides how different the four images in your starting grid are from one another. At 0 they look like small variations on one idea. Push it toward 100 and each of the four can wander somewhere surprising, which is great for exploring new directions. Default 0.

--weird / --w

Adds quirky, unusual, sometimes strange aesthetics you would not normally get. It ranges 0 to 3000; start small, around 250 to 500. It pairs nicely with --stylize, which keeps results attractive while --weird keeps them odd. Default 0.

--quality / --q

Sets how long Midjourney spends rendering, which affects detail and how many GPU minutes you use. It accepts .25, .5 and 1, with a default of 1. Lower values are faster and cheaper; higher spends more time but is not automatically "better," and flat or simple styles can look nearly the same at .5.

--seed

Every image starts from a field of random noise. The --seed parameter locks that starting point, using a number from 0 to 4294967295, so the same prompt, version and seed give you a repeatable, near-identical result. It is handy for making small tweaks without the whole picture changing.

--no : negative prompt

Tells Midjourney to try to leave something out. --no text helps drop letters and captions; --no plants clears greenery. It is a gentle nudge rather than a hard rule, so stubborn elements may still slip through. You can list several things, for example --no people, cars, signs.

--tile

Makes images whose edges line up seamlessly, so they repeat forever like wallpaper, fabric or a game texture. It is a simple on/off flag with no number attached. Combine it with a clean, evenly lit subject for the smoothest seamless result.

--style raw

Turns down Midjourney's automatic "make it pretty" instinct, giving you a more literal, photographic result that follows your prompt more exactly. It is great for realism and precise control. Write it as --style raw, available from v5.1 through v7. Niji has its own styles instead: cute, expressive, scenic and original.

--cref : character reference

Keeps the same character, the same face and overall look, across many images. Add --cref followed by the image URL of your character. Pair it with --cw (character weight, 0 to 100): --cw 100 copies face, hair and clothing, while --cw 0 focuses on just the face so you can change the outfit. It was introduced in v6. In v7 this role is handled by omni-reference, using --oref with --ow.

--sref : style reference

Copies the look and feel, the colors, texture and mood, of a reference image without copying its subject. Add --sref and an image URL, or a numeric style code. Control the strength with --sw (style weight, 0 to 1000, default 100). You can blend several references together to build a custom aesthetic, and --sref random hands you a surprise style to explore.

--iw : image weight

When you paste an image as part of your prompt alongside words, --iw decides how much that image matters versus the text. In v6 it ranges 0 to 3 with a default of 1; higher makes the output hug the reference image more closely, while lower lets the words lead. The exact range shifts between model versions.

Other handy parameters

Putting parameters together

Parameters stack at the very end of your prompt, each one starting with a double dash and separated by a space. Order does not matter, so you can add and remove them freely. A full prompt might read:

That asks for a widescreen image, drawn on version 6.1, in the more literal raw style, with a gentle push of Midjourney's own aesthetic, and no people in the scene. Change one number at a time and re-run, and you quickly learn what each dial actually does.

Free, copy-paste prompts to test them on

Parameters shape the frame, but the words inside still do most of the work. Start from a strong prompt instead of a blank box: