Parameters are the knobs at the end of a Midjourney prompt. They come after the description, each one starting with a double dash, and they control framing, how hard the model pushes its own aesthetic, how much it varies from your words, and which version does the work. Learn the handful that matter and you stop fighting the bot for the result you pictured.

This is a practical cheat sheet for Midjourney parameters. I will explain what each flag does, give the values worth memorizing, and show real prompts so you can see the syntax in context. The parameters themselves are evergreen — they have behaved consistently across model versions even as the numbers shift. Where a value is version-specific I will say so, and I will keep the explanations to behavior you can rely on.

How Midjourney parameters work #

Every parameter goes at the very end of the prompt, after all your descriptive text. The format is a double dash, the parameter name, a space, then the value. You can stack as many as you want:

/imagine prompt: a red fox in a snowy forest --ar 16:9 --stylize 250 --v 6.1

Order among the parameters does not matter — --ar 16:9 --v 6.1 and --v 6.1 --ar 16:9 produce the same job. What matters is that they sit after your description, not in the middle of it. Drop a parameter into the middle of your prompt and Midjourney may read the dashes as part of the text. Keep them at the end, always.

Most flags have a short alias and a long form. --ar is the same as --aspect; --s is the same as --stylize. Use whichever you remember. Below I lead with the form people actually type.

–ar (aspect ratio): the shape of the frame #

This is the parameter you will set most often. It controls the width-to-height ratio of the image. Without it, Midjourney defaults to a square.

/imagine prompt: a lone lighthouse on a rocky coast at sunset --ar 16:9 --v 6.1

Common ratios and when to reach for them:

  • --ar 1:1 — square. Social posts, logos, product grids, profile images.
  • --ar 16:9 — widescreen. Landscapes, desktop wallpaper, video thumbnails, hero banners.
  • --ar 4:5 — tall. Portraits and the standard mobile feed crop.
  • --ar 2:3 — taller still. Posters, book covers, full-body character art.
  • --ar 9:16 — vertical. Phone wallpaper, stories, reels.
  • --ar 21:9 — ultrawide. Cinematic stills and panoramic scenes.

The ratio does not just crop — it changes the composition. A castle at --ar 9:16 becomes a tall tower; the same castle at --ar 21:9 spreads into a sprawling fortress. Set the ratio before you fuss with anything else, because it shapes what Midjourney draws.

–stylize (–s): how hard Midjourney pushes its own taste #

Stylize controls how much of Midjourney’s trained aesthetic gets applied on top of your prompt. Low values stick close to what you wrote. High values let the model add its own color, lighting and flourish, sometimes drifting from your literal description.

/imagine prompt: a simple ceramic mug on a white table --ar 1:1 --stylize 50 --v 6.1
/imagine prompt: a simple ceramic mug on a white table --ar 1:1 --stylize 600 --v 6.1

Run that pair and the difference is obvious. The first is plain and faithful — close to a real product photo. The second is dramatic, with invented lighting and richer color that you never asked for. Neither is “correct”; they serve different jobs.

How I think about the range:

  • --stylize 0–100 — literal. Use for product shots, logos, diagrams, anything where accuracy beats flair.
  • --stylize 100–250 — balanced. A sensible default for most photographic and realistic work. The standard default sits in this zone.
  • --stylize 250–500 — expressive. Illustration, concept art, fantasy, editorial. Midjourney takes real creative liberties here.
  • --stylize 500–1000 — maximal. Abstract art and bold stylized pieces where you want the model to run wild.

Rule of thumb: the more precise your intent, the lower the stylize. If you need the product to match the real thing, keep it down. If you want a beautiful surprise, push it up.

–v (version): which model renders the job #

The version flag selects which Midjourney model handles your prompt. Newer versions read prompts more literally, render hands and text better, and handle complex scenes more coherently. As of this writing, --v 6.1 is the current-era general model.

/imagine prompt: a busy farmers market on a sunny morning, detailed --ar 3:2 --v 6.1

For anime and illustration there is a separate model line, Niji, which you invoke with --niji instead of --v. We will get to that below. The practical point: pick your model family first (general vs Niji), because it changes how every other parameter behaves. You can compare the two lines in detail in our Niji vs Midjourney breakdown.

–niji: the anime and illustration model #

Niji is Midjourney’s model tuned for anime, manga and illustration. Swap --v for --niji and the same prompt comes out with cleaner linework, flatter shading and anime proportions.

/imagine prompt: a teenage girl with twin braids standing in a flower field, soft lighting, anime style --ar 2:3 --niji 6

Niji pairs especially well with --style presets, which we cover next. If anime is your main interest, the dedicated anime AI art prompts guide goes deep on getting consistent character work out of Niji.

–style: presets that shift the whole look #

The style parameter applies a named aesthetic preset. Its options depend on the model. On Niji you have presets like --style raw, --style cute, --style scenic and --style original, each pulling the result in a different direction.

/imagine prompt: a knight resting by a campfire at night --ar 16:9 --niji 6 --style raw

--style raw is the one to know. On both the general and Niji models it reduces Midjourney’s automatic beautification, giving you a more neutral, photographic, less “Midjourney-looking” result. When you want the model to follow your prompt and stop adding its own gloss, raw is the lever:

/imagine prompt: candid street photo of an old bicycle leaning on a brick wall --ar 3:2 --style raw --v 6.1

–chaos (–c): how varied the four results are #

Every job returns a grid of options. Chaos controls how different those four are from each other. Low chaos gives four close cousins. High chaos gives four wildly different interpretations of the same prompt.

/imagine prompt: an abstract poster about the ocean --ar 2:3 --chaos 50 --v 6.1

The range runs 0 to 100. Reach for high chaos early, when you are exploring and want range. Drop to low chaos once you have a direction and want consistent variations to refine. I usually start a brand-new concept around --chaos 25 to --chaos 50, then go back to 0 once I know what I am after.

–no: negative prompting #

The --no parameter tells Midjourney to steer away from something. List the thing you do not want after it.

/imagine prompt: a bowl of fresh fruit on a wooden table --ar 1:1 --no bananas --v 6.1

It is not a hard guarantee — Midjourney reduces the weight of the named element rather than forbidding it outright — but it works well for removing unwanted objects, colors, or styles. Common uses: --no text to fight stray lettering, --no people to clear a landscape, --no blur to push for sharpness. For the full technique across models, see our guide to negative prompts.

–seed: reproducibility #

Every job uses a starting seed number. Supply a fixed seed and you get a repeatable starting point, which is useful when you want to change one word in a prompt and keep everything else close.

/imagine prompt: a cozy cabin in the snow --ar 16:9 --seed 1234 --v 6.1

Run that, then run it again with one descriptor changed and the same seed, and the two results stay closely related instead of diverging completely. This is the backbone of controlled iteration. It does not produce pixel-identical output across every change, but it anchors the composition so your edits are comparable.

–tile: seamless repeating patterns #

The tile parameter makes an image that repeats seamlessly on all edges — ideal for textures, fabrics, wrapping paper and web backgrounds.

/imagine prompt: watercolor floral pattern, soft pinks and greens --tile --v 6.1

Drop the result into any tiling tool and it repeats without visible seams. Pair it with --ar 1:1 and a low stylize for clean, predictable patterns.

–weird (–w): controlled strangeness #

Weird pushes the image toward the unusual and experimental without fully breaking it. The range runs 0 to 3000. Small amounts add an offbeat edge; large amounts get genuinely surreal.

/imagine prompt: a teapot designed by an alien civilization --ar 1:1 --weird 1000 --v 6.1

It is a fun way to escape generic results. I keep it under 500 when I still want the subject recognizable, and let it climb when I am chasing something strange on purpose.

Image prompts and –iw (image weight) #

You can put an image URL at the start of a prompt to use it as a visual reference. The --iw parameter then controls how strongly that image influences the result versus your text. Higher values lean on the image; lower values lean on the words.

/imagine prompt: https://example.com/reference.png a portrait in the same color palette --iw 1.5 --v 6.1

This is how you carry a mood or palette from one image into a new composition. Balance it against your text weight and tune --iw up or down until the blend feels right.

Putting parameters together #

You rarely use one flag in isolation. A real working prompt stacks several:

/imagine prompt: a misty mountain temple at dawn, golden light, atmospheric --ar 21:9 --stylize 300 --chaos 20 --no people --v 6.1

Read that back to front: render on version 6.1, with no people, only mild variation between the four results, expressive stylization, ultrawide cinematic frame. Each flag does one job, and together they pin the result close to what you imagined.

A few habits that save grief. Set --ar first, before you touch anything else, because it changes composition. Treat --stylize as your accuracy dial — down for precision, up for art. Use --seed the moment you want to iterate on a result you like. And keep every parameter at the end of the prompt, never in the middle.

Make parameters second nature #

You do not need to memorize all of these on day one. Start with --ar, --stylize and --v — those three cover the vast majority of jobs. Add --style raw, --no and --seed as you hit the situations that call for them. Once the flags are reflex, the rest of prompting gets easier, because you are no longer guessing at framing and intensity. To see parameters working inside a full prompt-building method, read our guide to Midjourney prompt structure, and use ArtPrompts Generator to assemble prompts with the right flags already attached.