AI image aspect ratios — pick the frame before you render

Every image render starts with a decision you cannot undo later: the shape of the frame. The Image studio offers five formats — one square, two portraits, two landscapes — and the model composes for whichever one is selected before it draws a single pixel. There is no crop step inside eroq and no upscaler, so the frame is a choice you make at the top of the job rather than a repair you make at the bottom.

The five formats and what they render at

Each ratio maps to a native resolution on the diffusion grid, which is why off-grid sizes are not offered at all:

• 1:1 — 1024 × 1024. The square. Profile pictures, feed posts, album art, model cards, anything that has to survive being displayed at 80 pixels wide.
• 3:4 — 896 × 1152. The standing portrait. People at full height, book covers, poster layouts, product packshots that are taller than they are wide.
• 4:3 — 1152 × 896. The classic landscape. Editorial photography, interiors, still life, anything that used to be shot on a camera rather than a phone.
• 9:16 — 768 × 1344. The phone frame. Story backgrounds, vertical ad stills, cover art for a short, anything that fills a screen held upright.
• 16:9 — 1344 × 768. Widescreen. Thumbnails, headers, hero images, and the start frame for a clip you plan to animate later.

All five cost the same: 10 credits an image on Image One, Image Anime and Image Art alike. There is no premium for a wide frame and no discount for a square one, which removes the only bad reason to pick the wrong shape.

A native ratio is not a crop

This is the part people learn the expensive way. If you render a square and then crop it to 16:9 in your editor, you get a square composition with its top and bottom removed — a head that was centered is now centered in a letterbox, and roughly 40% of the pixels you paid for are in the bin. Render at 16:9 and the model knows from the start that it has horizontal room: it places the subject off-center, lets the background run out to the sides, and puts the light where a wide frame wants it.

The same is true going the other way. A portrait is not a landscape stood on its end. Ask for 9:16 and the model stacks the scene — foreground at the bottom, subject in the middle third, sky or ceiling above — because that is what tall frames do. Ask for 16:9 and it spreads.

So the useful habit is to decide the destination first and set the format second, before you spend a word on the prompt. The aspect ratio glossary entry has the general definition; what matters here is that on eroq the ratio is a real parameter on the image models, not a hint you type into the prompt and hope for.

Choosing by destination

• Profile picture, avatar, app icon, album cover — 1:1. It is the only format that is never cropped by anyone.
• YouTube thumbnail, site hero, blog header, presentation slide — 16:9.
• TikTok and Reels stills, story backgrounds, vertical ads, phone wallpapers — 9:16.
• Book cover, poster, standing figure, tall packaging — 3:4.
• Editorial and documentary stills, interiors, food, anything with a horizon — 4:3.
• A still that will become a clip — match the video. The Ads templates in the studio use 9:16 for TikTok and Reels, 1:1 for feed, and 16:9 for YouTube and web, and an image-to-video source is easier to work with when it already has the shape of the clip.

Vertical formats deserve one extra note: the middle of a 9:16 frame is the only part guaranteed to be visible, because platforms put interface on the top and bottom. Compose the subject into the middle third and treat the rest as padding.

There is no 21:9 image and no upscaler

Two limits worth knowing before you plan around them.

Video takes six ratios on eroq, including 21:9 cinemascope. Images take five. If you want a cinemascope still, render 16:9 and trim it in your editor, accepting that you are cropping a frame the model composed for a different shape — or render the shot as a clip and pull a frame.

And there is no upscaling step anywhere in the product. What the model renders is what you get, at the native size above. Nobody is going to rescue a 768-pixel-tall render into a print file, so if the work needs height, pick a format that has it.

Write the prompt for the frame you chose

A prompt that knows its format is a better prompt. Name the framing in words as well as in the format pill — the two agree and reinforce each other, rather than the model having to guess what you meant by a wide canvas.

Wide establishing shot of a lone climber crossing a snowfield below a granite wall, figure small in the lower left third, the ridge line running out to the right, 35mm lens, hard low sun raking across the snow, blue shadow in the tracks behind her, vast and cold.

Read that in 16:9 and it works: the subject is placed, the space is used, the eye travels. Read it as a square and the instructions about the lower left third and the ridge running right have nowhere to go. That is the whole lesson in one paragraph.

When the format setting goes quiet

There is one case where the format pill does nothing at all: attaching a single reference photo. With exactly one reference the engine works from that picture rather than drawing a new one, and the source photo decides the frame — its shape is inherited, whatever the pill says. This is expected behavior rather than a bug, and it is covered properly in the image-to-image editing guide.

Batch, then remix

Since every format is the same price, the cheap way to settle an argument about framing is to run the same prompt twice in two shapes — 20 credits — and look at them side by side instead of imagining them. Batches of 1 to 4 render at the same settings for comparing takes within one format.

Every render is saved to your library with its full recipe, so Remix reloads the prompt and settings into the tool with the format ready to change. The seeds and remix workflow piece covers that loop; for the video side of the same question, see the video aspect ratios guide.

FAQ

Does a wider image cost more credits?

No. Every image is 10 credits on all three models, at any of the five formats. The resolution differs by ratio — 1024 × 1024 for a square, 1344 × 768 for widescreen — but the price does not, so pick the shape the work needs.

Can I change the aspect ratio of an image I already rendered?

Not inside eroq, because there is no crop or resize tool and no upscaler. What you can do is Remix the render, which reloads its prompt and settings, switch the format and render again for another 10 credits. That gives you a composition built for the new shape rather than a trimmed version of the old one.

Why is the format setting ignored when I attach a photo?

Because a single reference photo puts the render into edit mode, where the source image sets the frame. The engine is modifying a picture that already has a shape rather than composing a new one, so the aspect parameter has nothing to act on.

Is there a 21:9 image format?

No. 21:9 exists for video only; images run 1:1, 3:4, 4:3, 9:16 and 16:9. For a cinemascope still, render 16:9 and crop in your own editor, or render the shot as a clip and take a frame from it.

Pick the destination, set the format, then write the prompt — open the Image studio and start at the top.