Which aspect ratio to use, and why it changes the image

Aspect ratio changes the image because the model composes for the canvas it is given, not for a square it later trims. Ask for the same portrait prompt at 1:1 and at 9:16 and you do not get the same picture with different edges. You get a different shot: different framing distance, different amount of floor and ceiling, often a different pose. Treat ratio as part of the brief, decided before you generate, not as a crop you apply afterwards.

What actually changes

Three things shift when the canvas shape changes.

Framing distance. A tall canvas invites the model to step back and include more of the body. A wide canvas invites it to move in on the face or pull back into the environment, depending on the subject. The model is matching the conventions of every image it has seen at that shape.

What fills the empty space. A tall frame has vertical space above and below the subject that must contain something. The model will supply a ceiling, a sky, a floor, or a foreground object. A wide frame has horizontal space and supplies a second wall, a horizon, or negative space beside the subject. If you do not say what goes there, the model invents it.

Subject placement. Composition conventions differ by shape. Square frames centre. Wide frames tend toward off-centre placement with the subject looking into the space. Tall frames stack elements. These tendencies are learned from the source material and they are strong.

There is a practical consequence. If you crop a 16:9 generation down to 9:16, you do not get what you would have generated at 9:16. You get a narrow slice of a wide composition, with the subject too large, the environment cut off at the sides, and dead space where the model never intended any.

1:1 square

Square is neutral and it centres. It is the shape with the least compositional opinion, which makes it useful for anything that needs to be read as an object: product shots, icons, avatars, single-item illustrations, album-style artwork.

Use it for: catalogue and product images, avatars, thumbnails that need to survive a circular mask, quick tests of a new prompt before you commit to a final ratio.

4:5 portrait

The standard tall shape for social feeds, and the most useful portrait ratio for people. It is tall enough to include the body from roughly the knees up without pushing the face small, and it takes more vertical room on a phone screen than a square does.

At 4:5 a portrait prompt typically produces a three-quarter body frame. You get wardrobe, hands, and posture, which is why it suits fashion, traditional dress, and anything where the garment matters as much as the face.

Use it for: feed posts, lookbook images, portraits where clothing is part of the subject, editorial-style single images.

3:4 portrait

Slightly taller than 4:5, and the classic shape of a print photograph in portrait orientation. It reads as photographic rather than as social media, partly because it matches common print sizes and partly because it matches what many cameras produce when rotated.

Use it for: printed portraits, framed wall pieces, book and magazine page layouts, any image destined for paper.

9:16 vertical

The full-height phone screen. Stories, reels, shorts, vertical wallpapers, and any format that fills a handset edge to edge.

This is the most aggressive ratio in terms of what it does to a prompt. At 9:16 the model has a large amount of vertical space to fill, so it will usually give you a full-body figure with substantial floor and ceiling, or a landscape with a tall sky. If you want a face large in a 9:16 frame you have to say so, because the default is to step back.

Two habits help here. Say what occupies the top and bottom of the frame, because something will. And leave room for interface elements if the image is going into a stories or shorts format, where the top and bottom strips are covered by controls and captions. Put nothing you need to see in the outer fifth at either end.

Use it for: stories, reels, shorts, phone wallpapers, vertical video keyframes.

16:9 landscape

The widescreen shape. Video frames, thumbnails, presentation slides, website hero images, desktop wallpapers on many machines.

A wide frame asks for context. Generate a portrait at 16:9 and you will get the subject placed off-centre with a substantial environment beside them, because that is how wide images are usually composed. That is either what you want or a problem, and the way to control it is to describe the environment deliberately rather than leaving it open.

For thumbnails, remember the image will be viewed small. It needs one clear subject, high contrast, and nothing important near the edges, because platforms trim thumbnail frames differently.

Use it for: video thumbnails, slide backgrounds, site headers, desktop wallpapers, establishing shots and keyframes for clips.

3:2 landscape

The 35mm frame. Slightly less wide than 16:9, and it reads as a photograph rather than as a screen.

This is the ratio to reach for when you want something to look like it was shot rather than designed. Landscape photography, documentary-style scenes, street work, and environmental portraits all sit naturally at 3:2, because that is the shape most of the reference material was captured in.

Use it for: landscape and travel scenes, photojournalistic looks, prints in standard photo sizes, anything meant to pass as a real photograph.

2:3 portrait

The 35mm frame rotated. The tallest of the common photographic ratios, and noticeably narrower than 3:4.

It suits a single standing figure, a tall structure, or a vertical composition with layered depth, because the narrow width forces the model to stack elements rather than spread them. It is the default shape for posters and for most standard photo print sizes in portrait orientation.

Use it for: full-length figures, posters, book covers, prints, architectural verticals.

Choosing quickly

• Product or avatar, destination unknown: 1:1
• Person in a feed, garment matters: 4:5
• Portrait for print or a page layout: 3:4
• Full-screen phone content: 9:16
• Thumbnail, slide, or site header: 16:9
• Should look like a real photograph, horizontal: 3:2
• Full-length figure or a poster: 2:3

Two rules that save reruns

Set the ratio as a parameter, not in the prose. Writing vertical format or 9:16 inside the sentence asks the model to read a canvas instruction as scene content. Use the field or flag your tool provides. Prompts in the image library keep ratio as a structured field for that reason, so one prompt can be run at several shapes.

Decide the ratio before you write the prompt, and write for it. At 9:16, say what fills the top and the bottom. At 16:9, say what sits beside the subject. At 1:1, say how tightly the subject is framed, because the square will not tell the model for you.

Not every generator offers every ratio. Some accept a fixed list, others take arbitrary dimensions within a pixel budget. Check what a tool supports before planning around a shape; the tools directory lists that alongside the other limits.