MidjourneyParameters & settingsPrompting basics

Midjourney Parameters Explained: Every One in V8

By Marco Cavazzana · Updated · 9 min read
ShareEmail
Short answer

Midjourney parameters are settings added to the end of a prompt with two hyphens, such as --ar 16:9 or --s 250. They control what words cannot: the image's shape, how much styling Midjourney adds, how varied the grid is, how strongly image and style references count, and which model runs. In the current default version, V8.2, Raw is --raw and the quality parameter --q no longer applies.

This is a reference page: every parameter has its own section, so you can link straight to one. Values and defaults come from Midjourney's documentation for the current default model, V8.2, with notes where V7 or older versions behave differently. Midjourney changes quickly, so if a value is rejected, check the version you are using first.

How parameters work

  • Parameters go at the end of the prompt, after the description. Text after a parameter breaks it.
  • Leave a space before the two hyphens and no space between them: poppies --ar 2:3, not poppies--ar 2:3 or - - ar.
  • No punctuation in or after parameters. A trailing comma or full stop causes errors.
  • Order between parameters does not matter. Many have a short form: --stylize 250 and --s 250 are the same.
  • Most parameters can also be set as defaults in the settings panel; a parameter in the prompt overrides the default for that prompt. See Midjourney settings explained.

Watch out for phones and word processors that turn two hyphens into a dash. Retype them by hand if a parameter is ignored.

Quick reference table

ParameterWhat it doesValues (V8.2)
--ar / --aspectImage shapeWhole-number ratios, default 1:1
--s / --stylizeMidjourney's own styling0 to 1000, default 100
--rawTurns off default stylingNo value
--c / --chaosVariety within a grid0 to 100, default 0
--w / --weirdUnusual results0 to 3000, default 0
--iwWeight of an image prompt0 to 3, default 1
--q / --qualityGPU time per imageNot supported in V8.x; V7: 1, 2, 4
--srefStyle reference image or codeURL, code or random
--swStrength of the style reference0 to 1000, default 100
--svStyle reference versionSee the Style Reference docs
--p / --profilePersonalisation profileNo value, or a profile ID or code
--noThings to leave outWords, comma-separated
--seedFixes the starting noise0 to 4294967295
--tileSeamless repeating patternNo value
--v / --versionModel version8.2 (default), 8.1, 7, 6.1…
--nijiAnime and illustration model7 (current)
--hd / --sd2048 or 1024-pixel imagesNo value (V8.1 and later)
--editUses the Edit modelWith attached images
--draftCheap, fast draft imagesNo value
--r / --repeatRuns the prompt several times2 up to a plan limit
--fast / --relax / --turboGPU speed for this promptNo value
--stealth / --publicVisibility on the websiteNo value
--motion, --loop, --end, --bsVideo settingsSee the video section

--ar (aspect ratio)

Images are square by default. --ar 16:9 makes them wide, --ar 4:5 or --ar 9:16 tall. Use whole numbers only. Midjourney composes for the ratio you give it, so set it from the start instead of cropping later. The maximum is 14:1 for standard images and 4:1 for HD images. The ratio is not the pixel size; that depends on the version and on SD or HD. Our aspect ratio guide lists ratios by platform.

--s (stylize)

Stylize sets how much artistic freedom Midjourney takes. Low values follow your prompt closely; high values look more designed but can drift from your words. The range is 0 to 1000 and the default is 100. When personalisation is on, stylize also sets how strongly your profile is applied. Start at 50 for product and documentary work, 100 for general use and 250 to 500 for mood pieces.

--raw (formerly --raw)

Raw turns off the automatic styling Midjourney adds. With short prompts you get plainer, more photographic images; with detailed style prompts your words control the look. Midjourney's current documentation writes it as --raw, and it can also be switched on in the Model section of the settings. Older guides, including some of ours, show --raw, the form used with earlier versions; if one form is rejected, try the other. Raw also works on video prompts and with the Edit model, where it makes instructions more literal. More in Raw mode explained.

--c (chaos, or Variety)

Chaos makes the four images in a grid differ more from each other. It is called Variety on midjourney.com. The range is 0 to 100, default 0. Use 20 to 40 when you are looking for ideas and 0 once you know what you want; high values follow the prompt less closely.

--w (weird)

Weird pushes towards odd, unconventional images. The range is 0 to 3000, default 0. Midjourney describes it as experimental, says what counts as weird may change over time, and notes it is not fully compatible with seeds. Chaos varies the grid; weird changes the character of each image.

--iw (image weight)

When you add an image prompt (an image in the Image Prompt section of the Imagine bar, or an image URL at the start of the prompt), --iw sets how much it counts against your text. Higher values make the image influence the result more. In V8.1 and V7 the range is 0 to 3 with a default of 1; Niji 7 allows 0 to 2. Try --iw 0.5 when the image takes over and --iw 2 when it is ignored. Image weight applies to image prompts only, not to style references (use --sw) or Edit model references. See image prompts.

--q (quality)

Quality sets how much GPU time is spent on the first grid. In V7 the default is 1 and you can use 2 or 4 for two or four times the GPU time; V6 used 0.5, 1 and 2. Midjourney's version chart lists the quality parameter as not supported in V8.1 and V8.2. In the current versions the choice that changes detail and cost is SD or HD (--hd), and draft mode is the cheap option. Quality never affected variations, inpainting or upscales.

--sref (style reference)

A style reference copies the look of an image, its colours, medium, textures and lighting, without copying its objects or people. Drag an image into the Style Reference section of the Imagine bar, or add --sref with a URL or a numeric style code. --sref random picks a random code and converts it to a real code on submission. You can mix several codes and combine codes with images. Keep the text prompt about content, not instructions: "detailed portrait of a dog", not "the look of this image but a dog". Find codes in the Styles tab of the Explore page; our guide to finding sref codes explains how.

--sw (style weight)

Style weight sets how strongly the style reference applies, from 0 to 1000 with a default of 100. Midjourney notes that in V7 it has more effect with style codes than with images. It does not work with moodboards. Lower it when the reference overwhelms your subject; raise it when the look is not coming through.

--sv (style reference version)

Style reference systems change between model versions, and --sv selects which one a prompt uses, so old codes can keep their look. The available values differ for V7 and V6 and are listed in Midjourney's Style Reference article. It does not work with moodboards. Record the version and --sv next to any code you rely on.

--p (personalisation)

--p applies your default personalisation profiles; --p followed by a profile ID applies a specific one. Profiles are built by picking images you like on the Personalize page, and your Global V7 profile also works in V8.1 and V8.2. When you submit, the ID turns into a code that records the profile as it was at that moment, so the code is what to save. Stylize controls how strongly personalisation is applied.

--no (negative prompt)

--no lists what to leave out: still life gouache painting --no fruit, apple, pear. Writing "without fruit" in the prompt tends to add fruit. Midjourney reads every word after --no separately, so --no modern clothing is read as "no modern" and "no clothing" and can trigger moderation. Describe what you want instead in cases like that. Details in negative prompts.

--seed

A seed fixes the starting noise, a whole number from 0 to 4294967295. With the same prompt, seed and settings, results come out very close: Midjourney describes V8.x seeds as 99% identical. Use seeds to test one change at a time. They do not store a style or a character, they can behave differently after a break or a model update, and they are unreliable in Turbo mode. Copy an image's seed from the options next to its prompt.

--tile

--tile makes the edges of the image match so it repeats seamlessly, for fabric, wallpaper and textures. It produces one tile; click the image twice on the website to preview the pattern. Midjourney advises against upscaling tiles because it often breaks the seams, and the Edit model is not compatible with --tile.

--v and --niji (version)

--v picks the model. V8.2 has been the default since 24 July 2026; V8.1, V7 and V6.1 remain available, and some features exist only on older versions, such as multi-prompts with :: (V6.1 and earlier) and Omni Reference (V7). --niji 7 uses the anime and illustration model made with Spellbrush. Fix the version per project: the same prompt looks different across versions.

--hd and --sd

From V8.1, --hd generates images at about 2048 pixels on the long side and --sd at the standard 1024. Midjourney lists HD at about 1.3 GPU minutes per prompt against 0.8 for SD. HD images cannot be upscaled further, and edits with Pan, Zoom Out or Vary Region return SD results. See how to upscale.

--edit

--edit runs the Edit model: change an image with written instructions or build a new one from up to four reference images. It replaces Omni Reference (--oref) and Character Reference (--cref) in V8.x. It matches the ratio of the first attached image unless you add --ar. Full walkthrough in the Edit model guide.

--draft, --repeat and speed

--draft runs a prompt in Draft mode. In V8.1 and V8.2 a draft prompt makes a batch of 24 small images (512 pixels) for about 0.4 GPU minutes, on the website only; in V7 it makes four images at half the usual cost. --r runs a prompt several times: Basic allows 2 to 4, Standard 2 to 10, Pro and Mega 2 to 40, in Fast and Turbo modes only. --fast, --relax and --turbo choose the GPU speed for one prompt; Turbo is not currently supported in V8.1.

--stealth and --public

These set whether a creation appears on the public website. Stealth is available on Pro and Mega plans. Images made in public Discord channels are visible regardless.

Video parameters

Video prompts only accept video parameters: --motion low (default, calmer) or --motion high (bigger movement, more glitches), --raw, --loop (the end frame matches the start), --end (a different end frame) and --bs for batch size 1, 2 or 4. Videos start at 5 seconds and can be extended by 4 seconds at a time to 21 seconds.

Legacy parameters

Several parameters from older guides no longer apply to current versions: --cref and --cw (V6 only), --oref and --ow (V7), --stop (V6 and earlier), --style codes from the retired Style Tuner, and multi-prompt weights with ::, which Midjourney lists for V6.1 and earlier. --exp appears in the version chart for V7 and V8.x but has no article of its own; treat it as experimental.

Try it

a lighthouse on a rocky coast at dusk, long exposure, moody blue light, 24mm landscape photograph --ar 3:2 --raw --s 75 --c 10

Change one parameter per run: first --s 400, then --c 40, then add --hd. Seeing the same scene move with each parameter teaches more than any table.

Frequently asked questions

What does --iw do in Midjourney?
It sets how strongly an image prompt influences the result compared with your text. In V8.1 and V7 the range is 0 to 3 with a default of 1, and Niji 7 allows 0 to 2. It affects image prompts only; style references use --sw instead.
Does --q still work in Midjourney V8?
No. Midjourney's version chart lists the quality parameter as supported in V6 and V7 only. In V8.1 and V8.2 the choice that affects detail and GPU cost is SD or HD, and Draft mode is the cheap way to explore.
Is it --raw or --raw?
Midjourney's current documentation uses --raw, and you can also switch Raw on in the settings panel. --raw is the form used with earlier versions, which is why many guides still show it. Both describe the same idea: turning off Midjourney's default styling.

Nothing you makeis lost.

Free · No card · Read-only · 2-minute setup

DreamdriveFree plan · Chrome
Add to Chrome — it’s free