The single most common misunderstanding about aspect ratio: people treat it as a delivery format decided at the end. It isn't. The model composes for the frame it's given, so ratio is one of the first decisions, and it silently rewrites everything downstream — subject scale, how much environment exists, where the eye lands.
What each frame tends to produce
| Frame | What the model tends to do | Good for |
|---|---|---|
| Portrait (2:3, 3:4, 9:16) | Fills the height with the subject; minimal environment; tight headroom; strong central vertical | People, single objects, product shots, anything for phones or Pinterest |
| Square (1:1) | Centred, balanced, subject at medium scale with a readable but secondary background | Feed images, thumbnails, icons, catalogue grids |
| Landscape (3:2, 16:9) | Subject placed within a setting; genuine negative space; horizon-driven composition | Scenes, environments, banners, anything where context is the point |
Try the same subject in two frames and the difference is not subtle. 'A lone figure on a salt flat' in portrait gives you a person, with a bit of salt flat behind them. The same words in 16:9 gives you a salt flat, with a small person in it, and the emptiness becomes the subject. Neither is more correct — but if you wanted the second and asked in portrait, no amount of prompt rewriting will fully recover it.
Match your camera language to the frame
Frame and lens language reinforce each other, and they can also fight. Asking for a 'wide establishing shot' in a 9:16 frame gives the model contradictory instructions: the words want breadth, the canvas wants height. You typically get a compromise that reads as neither — a cramped landscape, or a vertical image with a strangely squeezed horizon.
- —Portrait pairs with: close-up, medium shot, 85mm, shallow depth of field, low angle, subject fills the frame.
- —Landscape pairs with: establishing shot, wide angle, 24mm, deep focus, rule-of-thirds placement, layered foreground and background.
- —Square pairs with: centred composition, flat lay, symmetrical framing, top-down.
Don't fix framing with a crop
Generating wide and cropping to portrait afterwards costs you resolution in the dimension you kept, and leaves a composition designed for a frame you're no longer using — the subject sits where it made sense in the wide version, which usually means dead-centre and too small. Regenerate in the target frame instead. It's cheaper than it sounds and the result is composed rather than salvaged.
How this shows up on the site
Every prompt page here records the orientation its image was generated at, right in the parameter table, and the composition of the resulting image is classified separately — centred, rule of thirds, symmetrical, and so on. Those are two different things and it's worth seeing them apart: the same portrait frame can hold a tightly centred subject or an off-centre one, and the composition pages let you browse by the result rather than the container.
- Browse by composition — Filter the gallery by how images are actually framed
- Browse everything — Full gallery with facet filters
- Browse by model — See the same kind of prompt across all three engines we publish with