Midjourney Commands & Prompt Reference

46 commands & parameters · 8 categories · Sources: Midjourney Docs — Parameter List (article updated 2026-09-01) and 28 more · Checked

Midjourney is driven by typed input: parameters at the end of a prompt set shape, style, references, GPU speed and video options, and Discord slash commands start or manage jobs. Checked against Midjourney's documentation as of October 2026, where V8.2 is the default model (since July 24, 2026). Notes flag parameters tied to certain versions, such as --edit in V8.1 and V8.2 or --oref in V7. Legacy parameters, grid buttons and website-only controls are left out.

All Commands

Composition and variety (7)

Command / SyntaxWhat it doesNotes
--arSet the aspect ratioLong form --aspect; write whole numbers such as 16:9 or 139:100, never decimals, and the default is a 1:1 square.
--chaosVary the four resultsShort form --c; values run from 0 (default) to 100, and the website calls this setting Variety.
--noExclude what you don't wantList several items separated by commas; moderation reads each word on its own, and it acts like a -0.5 multi-prompt weight.
--seedFix the starting noiseTakes a whole number from 0 to 4294967295; in V8.x the same seed and prompt give about 99% identical images.
--tileSeamless repeating tileProduces a single tile whose edges join up; upscaling it usually breaks the seamless pattern.
--repeatRun a prompt several timesShort form --r; Fast and Turbo only, up to 4, 10 or 40 runs by plan, and it is stripped from the finished prompt.
--weirdAdd unusual aestheticsShort form --w; 0 by default and up to 3000, works in V5 and later but does not combine reliably with seeds.

Style and personalization (7)

Command / SyntaxWhat it doesNotes
--stylizeStrength of the house styleShort form --s; 0 to 1000 with 100 as default: low values stay close to the prompt, high values interpret it freely.
--rawRaw modeSwitches off the automatic styling, so simple prompts come out more photographic; available from V5.1 onward and on video prompts.
--profileApply PersonalizationShort form --p; adds your default profiles or moodboards, or one specific profile when followed by its ID.
--srefStyle ReferenceTakes one or more image URLs or numeric style codes and copies colors, medium and lighting, not objects; V6 and later.
--sref randomRandom style codeThe word is replaced by a real code after submission, so rerolls keep that look; in V7 it needs --sv 4 or 6.
--swStyle WeightRanges from 0 to 1000 with 100 as default; it cannot be combined with Moodboards.
--svStyle reference versionV7 has six versions with 6 as default and 4 as the pre-June 2025 model; V6 has four, defaulting to 4.

Image references and editing (4)

Command / SyntaxWhat it doesNotes
--editEdit model referencesV8.1 and V8.2 only: up to four image URLs plus plain-language instructions; replaces Omni and Character Reference.
--iwImage WeightDefault 1; range 0 to 3 in V7 and V8.1, 0 to 2 in Niji 7; higher values follow the image prompt more closely.
--orefOmni ReferenceV7 only, one image per prompt at twice the usual GPU time; in V8.1 and V8.2 use --edit instead.
--owOmni Reference weightV7 only; accepts 1 to 1000 with 100 as default, and weights above about 400 turn unpredictable unless stylize is very high.

Model version and quality (6)

Command / SyntaxWhat it doesNotes
--versionChoose a model versionShort form --v plus a number; without it, prompts use your settings default or V8.2, the default since July 24, 2026.
--nijiNiji anime modelWrite --niji 7 for Niji 7, launched January 2026 with Spellbrush, or --niji 6 for the model that pairs with V6.
--draftDraft ModeIn V7 it gives four images at half cost; in V8.1 and V8.2 it gives 24 images at 512px, on the website only.
--qualityRendering qualityShort form --q; V7 accepts 1, 2 or 4 and V6 accepts 0.5, 1 or 2; V8.1 and V8.2 lack it.
--hdHD images at 2048pxV8.1 and V8.2 only; costs 1.3 GPU minutes instead of 0.8 and limits the aspect ratio to 4:1.
--sdStandard images at 1024pxThe cheaper V8.1 and V8.2 size; Pan, Zoom Out or Vary Region results from HD images also come out at SD.

GPU speed and visibility (5)

Command / SyntaxWhat it doesNotes
--fastFast mode for one promptSpends monthly Fast time, which resets at renewal; in Discord the standalone /fast command changes the default instead.
--relaxRelax mode for one promptStandard, Pro and Mega plans; uses no Fast time but may queue up to 30 minutes, and repeat or permutations will not run.
--turboTurbo mode for one promptUp to four times faster than Fast but charges double Fast time; not supported in V8.1 and V8.2.
--stealthKeep results privatePro and Mega plans only; images made in public Discord channels remain visible to others there.
--publicPublish results on the websiteOverrides Stealth for one prompt; /stealth and /public in Discord switch the default mode instead.

Video (6)

Command / SyntaxWhat it doesNotes
--videoAnimate an image in DiscordPaste the start image URL first; text is optional and only video parameters apply to the 5-second clip.
--motion lowLow motionThe default; favours still scenes, slow camera work and subtle character movement.
--motion highHigh motionLarger camera and character movement, at the risk of unrealistic or glitchy motion.
--loopLoop the videoReuses the first frame as the last one, so the clip ends where it began.
--endSet an end frameFollow it with an online image URL; a file on your device has to be hosted first, for example in Discord.
--bsVideo batch sizeNumber of videos per prompt: 1, 2 or 4 (default 4); a single SD video costs 2 GPU minutes instead of 8.

Prompt syntax (2)

Command / SyntaxWhat it doesNotes
::Multi-prompt separatorA number after it sets that part's weight (space::2 ship); works up to V6.1 and Niji 6, not in V7 or V8.
{}Permutation optionsComma-separated choices in the braces expand into separate jobs, as in --ar {1:1, 2:3}; Fast or Turbo only, up to 40 by plan.

Discord commands (9)

Command / SyntaxWhat it doesNotes
/imagineGenerate from a promptWrite the description first and any parameters last, separated by a space and with no punctuation inside them.
/describePrompts from an imageUpload a file or give an image link; four suggestions come back, each runnable from its numbered button.
/blendBlend two to five imagesIgnores text; output is square unless the dimensions field asks for 2:3 portrait or 3:2 landscape.
/settingsDefault settings panelSets the default version, Raw, stylize presets and more; only you see the panel and it syncs with the website.
/infoAccount and job infoShows remaining Fast time, plan, renewal and visibility mode in a panel that only you can see.
/showRepost a job by IDPaste a Job ID into the job_id field to bring one of your own images back into the current channel.
/prefer suffixAppend text to every promptRunning it again without new_value clears the suffix; the current suffix shows in the settings panel.
/prefer option setSave a parameter bundleA set named mine is then typed as --mine in prompts; sets are created only in Discord but also work on the website.
/prefer remixToggle Remix modeWith Remix on, the prompt can be edited when making variations, rerolling, panning or using Vary Region.

Recipes

Combinations of the parameters above, each checked against the Midjourney documentation for V8.2 and V7.

Carry one look across a whole set

--sref random

After the job is submitted the word random becomes a numeric style code. Copy that code into later prompts after --sref and tune its strength with --sw (0 to 1000, default 100).

Put the same person or object into a new scene

--edit

In V8.1 and V8.2, add up to four image URLs after --edit and describe the change in plain words; --raw gives the text more weight. In V7 the equivalent is --oref with a weight set by --ow.

Follow the words more literally

--stylize 50 --raw

A low stylize value keeps the image close to the prompt (50 is the Stylize Low preset in Discord settings), and --raw removes the default styling on top of that.

Try several aspect ratios in one go

--ar {1:1, 2:3, 3:5}

A permutation expands into one job per option. It needs Fast or Turbo mode, and each job spends Fast time.

Make test videos cheaply

--video --bs 1

One SD video costs 2 GPU minutes against 8 for the default batch of four. Keep the default --motion low unless the scene needs large camera moves.

Reuse your usual parameters

/prefer option set

Save a bundle such as chaos plus aspect ratio under a name like mine, then type --mine at the end of any prompt. For parameters you want on every prompt, /prefer suffix adds them automatically.

Frequently Asked Questions

Where do parameters go in a Midjourney prompt?

At the very end, after the description, with a space before the two dashes and no commas or periods inside them. Text typed after a parameter is one of the formatting mistakes the documentation warns about.

What replaced Omni Reference and Character Reference in V8?

The Edit model. In V8.1 and V8.2, --edit followed by up to four image URLs carries people or objects into new images; --oref and --ow still work, but only in V7.

Which model does a prompt use when no version is given?

V8.2, the default since July 24, 2026, unless another default was chosen in settings. Add --version (short form --v) with a number, such as 7, to use an older model.

Which parameters stop working in V8.1 and V8.2?

The version chart marks --quality, Turbo mode, Omni Reference (--oref, --ow) and multi-prompts with :: as unavailable there, while HD output with --hd and --edit exist only in those versions.

Sources and checks

Every row on this page was matched to a line in the documentation above on . A command or parameter that could not be found there is not listed.

2 rows from the earlier version of this page were removed in that check: duplicates, and entries the documentation does not list or lists differently.

AI tools change their commands, parameters and model names between releases: the vendor's current documentation and the version you run have the last word. Found a wrong or missing entry? Tell us and it gets checked against the source. How we verify describes the process and its limits.

Related prompt references

Kling AI (54) · Vidu (49) · MiniMax (Hailuo) (72) · All AI prompt references