Choosing --ar: what 85 images say about frame and subject

--ar looks like the least interesting parameter in Midjourney. It sets the shape of the output; you presumably already know what shape you need. But the model does not treat the frame as a crop applied at the end — it composes into the frame. Ask for 16:9 and you get a picture built for width: more setting, more horizon, the subject placed rather than centred.

Which means the ratio you pick is a compositional instruction, and the way to learn it is to look at what ratio real prompts use for real subjects.

• 50of 85 images set --ar explicitly
• 34 of 35prompts with no --ar come out square
• 9distinct ratios in use
• 7 of 9architecture images are 16:9

The default is a decision too

Thirty-five images here carry no --ar at all, and thirty-four of them come out square. Meanwhile not one of the fifty prompts that do set --ar asks for 1:1 — every square in the collection is a default nobody overrode. That is worth noticing before anything else: if you never think about the frame, you have chosen 1:1, and 1:1 is a strong choice. It centres. It flattens depth. It suits a single object and fights a landscape.

Architecture takes the wide frame

Seven of the nine architecture images use --ar 16:9. That is the strongest correlation between a parameter and a subject anywhere in this collection, and it is not a stylistic preference — buildings need somewhere to stand. A square crops the context that makes a structure legible; a wide frame gives you approach, ground and sky.

Six of the seven 16:9 architecture images. The frame is doing structural work: in every one of them the building is read against its surroundings rather than isolated.

The architecture prompts are also the shortest in the collection — an average of 45 characters of description against 170 for the photorealistic ones. None of them use --raw, none use --niji, and only two carry a style code. A short description, a wide frame, and nothing else:

Game concept art of an alien world, exotic vegetation, twin suns in the sky, atmospheric perspective, detailed environment design 

That combination works because the subject is doing the heavy lifting. You do not need to describe the light on a building the way you need to describe the mood of a portrait.

3:2 is the photograph

Six images use --ar 3:2, and five of the six are photorealistic. 3:2 is the 35mm frame — the shape of essentially every photograph taken on a film SLR or a full-frame digital body. Asking for it is a way of saying "this is a photograph" without spending words on it.

The five 3:2 photographs. They also happen to share a single style code and a --stylize 50 --raw tail — the frame is one part of a consistent photographic setup.

Tall frames, and the one odd ratio

Fifteen images use a portrait ratio — six at 3:4, six at 4:5, three at 2:3. Almost all are concept art. The distinction between them matters more than it looks: 4:5 is the social-feed frame, 3:4 is the classical portrait, 2:3 is the poster.

And then there is one image at --ar 58:77, which is not a standard anything. Midjourney accepts arbitrary ratios, and occasionally that is the right answer — matching a physical print size, or a layout you already have. Worth knowing the door is open, even if you rarely walk through it.

Practical rules

• Set it before you iterate, not after. Changing --ar re-composes the image rather than cropping it, so a ratio change late in a run costs you the composition you were tuning.
• Wide for context, square for object, tall for figure. That is what the distribution above says, stated plainly.
• Leaving it out is choosing 1:1. Fine when the subject is a single thing. Costly when it is a place.
• Non-standard ratios are allowed. If you have a real target — a cover, a print, a banner — ask for that ratio directly rather than generating 16:9 and cropping.

Every image referenced above links to its own page with the full prompt, so you can check any of these claims against the actual parameters rather than taking them on trust.