parameters — --ar

Setting the frame with --ar

Every prompt without --ar is a prompt asking for a square. The syntax has three rules worth knowing and one common confusion with image resolution.

--ar sets the shape of the frame — width against height, written as two numbers with a colon. It is the most consequential parameter that people forget, because leaving it out is not a neutral act: Midjourney images are square by default.

--ar <width>:<height>
--aspect <width>:<height>    (same parameter, long form)
  • 1:1the default — a square
  • 14:1widest supported (4:1 on HD)
  • 0decimals allowed

Reading the notation

Bigger first number, wider image. Bigger second number, taller image. That is the whole rule, and the common ratios are worth memorising because they map to real destinations:

RatioShapeWhere it goes
1:1squareprofile pictures, feed posts
4:3slightly wideolder screens and monitors
2:3tallprinted photography, picture frames
16:9widescreenHD video, modern TVs, desktop
9:16verticalphone screens, stories and reels

Three syntax rules

RuleWrongRight
No decimals--ar 1.39:1--ar 139:100
Whole numbers, either size--ar 1920:1080
No punctuation after it--ar 16:9,--ar 16:9

The second row is a genuinely useful trick rather than a curiosity. You do not have to reduce the fraction yourself — pass the dimensions and Midjourney simplifies --ar 1920:1080 to 16:9 for you. For paper sizes, strip the decimal point: US Letter at 8.5 × 11 inches becomes --ar 85:110.

Where it stops working

ConstraintDetail
Maximum ratio14:1 on V6, V7, V8.1 and V8.2 — but 4:1 for HD images
Extreme ratiosvery wide and very tall are experimental and unpredictable
Upscalingsome ratios shift slightly when upscaled
Older versionsmay not support the full range

The HD asymmetry is the one to plan around: if a panorama is the goal, HD's 4:1 ceiling arrives long before the standard 14:1 does, so the resolution decision and the frame decision are not independent.

Choosing one, briefly

Pick the ratio your image has to live in, not the one that looks best in a generation grid. A vertical crop of a wide image is a worse picture than a vertical image; the frame changes composition, not just edges.

And if you get it wrong, it is recoverable without regenerating: pan and zoom-out extend an existing image, and the Editor can change a finished image's proportions outright.

The collection records each image's ratio as a field of its own, so you can filter to a shape and see what people do with it.