Midjourney parameters explained: --ar, --v, --style raw and friends

A Midjourney prompt has two parts: the description and the parameters. The description tells Midjourney what to make; the parameters, everything that starts with a double dash, tell it how. Mixing them up is one of the most common reasons results don’t match what you had in mind.

Parameters always go at the very end of the prompt, separated by spaces. Anything after the first double dash is read as settings, not description.

--ar: aspect ratio

--ar 3:2 sets the shape of the image. The default is square (1:1). Choose the ratio before anything else, because it changes composition: a portrait subject in a wide 16:9 frame gets surrounded by background, while the same subject in 2:3 fills the frame.

Common choices are 3:2 for classic photos, 16:9 for wallpapers and thumbnails, 9:16 for phone screens and stories, and 4:5 for social feeds. When you use our Midjourney prompt generator, the ratio is taken from your uploaded image automatically, so the result keeps the original framing.

--v: model version

--v 7 selects the model version. Different versions interpret the same words differently, so fixing the version makes results reproducible and lets you share prompts with others who will get similar output. When a new version comes out, test your favourite prompts with it before switching everything over.

--style raw: less house style

Midjourney applies its own aesthetic by default: richer colours, more drama, more polish. --style raw turns much of that down and makes the model follow your words more literally. Use it when you are matching a photograph or need a specific, understated look. Leave it out when you want Midjourney’s characteristic painterly finish.

--stylize: how strongly the aesthetic is applied

--stylize (or --s) takes a number, typically from 0 to 1000. Low values stick closely to the prompt; high values give Midjourney more artistic freedom. If your results look beautiful but ignore half of what you asked for, lower the value. If they look accurate but flat, raise it.

--chaos: how varied the four results are

--chaos controls how different the four images in a grid are from each other. At 0 they are close variations; at higher values they explore very different interpretations. Raise it while you are exploring ideas, and lower it again once you know what you want.

--no: leave something out

Midjourney doesn’t reliably understand negation in the description. “A street without cars” can easily produce cars, because the word “cars” is in the prompt. Use --no cars instead. It works for objects, colours and styles, and you can list several items separated by commas.

--sref: borrow a style

--sref followed by an image URL uses that image as a style reference: colours, lighting and texture, without copying its content. It is the most effective way to keep a consistent look across a series. Combine it with a prompt from our generator to describe the new subject in the style of your reference.

Putting it together

A complete prompt might look like this:

a lone lighthouse on a basalt cliff above a stormy coastline, shafts of golden light through heavy clouds, painterly matte painting, slate blue and burnished gold --ar 3:2 --style raw --stylize 150 --v 7

Description first, most important words at the start, parameters at the end. The same structure works for almost any subject.

Mistakes to avoid

• Putting parameters in the middle. Everything after the first -- is treated as settings, so description words after it may be ignored.
• Contradicting yourself. “Photorealistic” with a high --stylize value pulls in two directions.
• Changing everything at once. Change one parameter per run so you can tell what made the difference.

For the descriptive part of the prompt, the habits in why your AI images look generic apply to Midjourney as much as to any other model. And if you want to start from an image you already like, reverse-engineering it into a prompt is the quickest route.