AI Image Sizes and Aspect Ratios: Cheat Sheet

AI image sizes cause more wasted regenerations than any prompt mistake: you get a lovely picture in the wrong shape, crop it for a story or a banner and lose the subject. The fix is to decide the format before you generate. Pick the aspect ratio for where the image will be shown, ask for that shape in the prompt, leave safe margins for cropping and check the actual pixel dimensions of the file you receive.

This cheat sheet covers the ratios that matter for social posts, link previews, video thumbnails, websites, e-mail and print, how to calculate pixels from print sizes, how to crop without ruining a composition and what to do when a generated image comes out at a different size. Platform specifications change, so treat the figures here as common starting points and confirm the current numbers in each platform’s help pages.

AI image sizes: aspect ratio versus pixels

An aspect ratio describes shape: width divided by height. A 1:1 image is square, 16:9 is wide like a video frame and 9:16 is the same shape turned upright. Pixel size describes how much detail the image holds. Two images can share a ratio and differ in sharpness, and two can share a pixel count and differ in shape. The aspect ratio article gives the background.

The useful rule is that you choose the ratio first and the pixel size second. A platform will usually accept a range of pixel sizes and crop or scale them, but it cannot add detail that is missing, and it will crop an image of the wrong shape in a way you do not control. When you generate with AI, the ratio is the thing to specify, and size is something you can often adjust afterwards.

To work out a missing dimension, multiply. For a 4:5 portrait that is 1080 pixels wide, the height is 1080 times 5 divided by 4, which is 1350. For a 16:9 image 1920 wide, the height is 1920 times 9 divided by 16, which is 1080.

Common ratios and where they fit

Ratio	Typical pixels	Common uses
1:1 square	1080 × 1080	Feed posts, profile pictures, product tiles
4:5 portrait	1080 × 1350	Feed posts that take more screen height on phones
9:16 vertical	1080 × 1920	Stories, short vertical video covers
16:9 wide	1920 × 1080	Video frames, presentation slides, website hero images
16:9 thumbnail	1280 × 720	Video thumbnails
About 1.91:1	1200 × 630	Link previews in social posts and messengers
2:3 portrait	1000 × 1500	Pinned images and tall visual posts
3:2 landscape	1800 × 1200	Blog images, photography, print photos

These are widely used values, not guarantees. Platforms update layouts and sometimes change their preferred sizes. Before a campaign, open the platform’s current help page, or post a test image and look at it on a phone.

Link previews and Open Graph images

When you share a web page, platforms show a preview built from the page’s Open Graph image. The Open Graph protocol defines the tags, and a size of 1200 by 630 pixels, roughly 1.91:1, is the common choice because it displays well across social networks and messengers. Keep important content, such as a title or face, away from the edges, since some platforms crop to a square or a different ratio in certain views.

Use a file that is not huge: a few hundred kilobytes loads quickly, and some services reject very large images. Save it as JPEG for photographs or PNG for flat graphics with text. Our guide to designing a social media banner with AI goes deeper on layouts for these formats.

Video thumbnails and wide images

Thumbnails are small and viewed on phones, so what matters is a clear subject and large, short text. The common 1280 by 720 size at 16:9 gives enough room for a face or an object and three to five words. Test your thumbnail by shrinking it to the size of a postage stamp: if you cannot tell what it shows, simplify. See AI YouTube thumbnail design for what tends to earn the click.

For website hero images, 16:9 or 3:2 are the usual shapes, but responsive layouts crop them. Leave empty space where text will sit, and keep the subject in the middle third so it survives different screen shapes.

Stories, Reels and vertical formats

Vertical formats at 9:16 are shown full screen, and interface elements cover parts of the image: the account name and progress bar at the top, and reply or caption controls at the bottom. Keep faces, logos and text out of roughly the top and bottom seventh of the frame. Ask for that in the prompt: “vertical 9:16 composition, main subject in the middle, empty space at the top and bottom.”

A feed post in 4:5 occupies more of the screen than a square, which is why many brands prefer it for photos. If you plan to reuse one image across formats, generate it wider or taller than needed and plan the crops, which the next section covers.

Generate once, crop many times

A single generation can serve several formats if you plan for it. Ask for a composition with generous background around the subject, then crop to each ratio. A subject placed in the centre of a 3:2 image survives a crop to 1:1, 4:5 and, with care, 16:9. A subject on the far left does not.

Pick the widest and tallest need. If you need both 16:9 and 9:16, decide whether one image can do both, or generate two.
Centre the subject and keep key details in the middle 60 percent.
Ask for plain background areas where text can sit.
Crop, do not stretch. Changing the ratio by scaling distorts faces and logos.
Export each format separately and name the files clearly, for example hero-16×9.jpg and story-9×16.jpg.

Our guide to consistent AI images shows how to keep one look across all the formats of a campaign.

Pixels for print

Print needs more pixels than screens. The calculation is simple: pixels equal inches times resolution in dots per inch. A common target is 300 dpi for sharp print, and 150 dpi can be enough for large items viewed at a distance.

Print item	Size	Pixels at 300 dpi
Business card	3.5 × 2 in	1050 × 600
Photo print	4 × 6 in	1200 × 1800
Greeting card front	5 × 7 in	1500 × 2100
A4 page	210 × 297 mm	about 2480 × 3508
A3 poster at 150 dpi	297 × 420 mm	about 1754 × 2480

Printers also ask for bleed, an extra margin of a few millimetres that is trimmed off, so check your printer’s template. Generated images may not reach print resolution directly. If yours is smaller than the table suggests, use it for layouts and drafts, or upscale carefully and inspect the result, since upscaling can invent unwanted detail. Related guides: AI poster maker and AI business card design.

When the image comes back the wrong size

Generated images do not always match the requested size exactly. The reliable habit is to check the file, not the preview. Open the properties of the downloaded file and read its width and height. If it is not what you need:

Close, but different ratio. Crop to the target ratio, keeping the subject.
Too small. Use it at a smaller display size, or upscale with care and check for artefacts.
Wrong orientation. Regenerate with an explicit “vertical” or “horizontal” instruction, and describe the composition for that shape.
Text garbled. Remove the text in the prompt and add it afterwards in an editor, where you control font and spelling. Our list of common AI image failures covers this and similar problems.

Never stretch an image to fit. A small crop is almost always better than a distorted whole.

File formats and weight

Choose the format by content. JPEG suits photographs and gives small files. PNG suits graphics with text, flat colours and transparency but makes larger files. WebP offers smaller files at similar quality and is widely supported on the web. For social uploads, platforms re-compress images anyway, so upload a clean, high-quality file within their size limits. For websites, resize to the largest size actually displayed, then compress, so that pages load quickly on mobile networks. Keep the originals in a folder, because you will need the full-size version the next time a layout changes.

Accessibility counts as well. Add a short alternative text to every image that carries information, describing what it shows in a sentence, and avoid putting essential words only inside the picture, since screen readers cannot read them and search engines may miss them. A caption or a nearby heading repeats the message in text and costs nothing.

Using Mio to plan image sizes

Ask Mio can create images in Design mode, which comes with the Design and Business plans, at 20 points per generated image, with an allowance of images per month on each plan. In your prompt, describe the shape you need, such as “vertical 9:16, main subject centred, empty space at top and bottom”, and the use, such as “for a story”. Mio routes each request to the model it judges best for the task, so you do not pick one. Always check the dimensions of the file you receive and crop or resize it in an editor if needed.

Chat and Write modes, available on every plan including the free one, are useful for planning. Ask Mio to turn a list of your channels into a table of ratios, pixel sizes and safe zones, or to write a short checklist for your designer. Our guides to AI image prompts that work and the AI image generator guide cover the prompts themselves.

Common mistakes

Choosing the ratio after generating. The crop then removes the subject.
Stretching to fit. It distorts faces and logos.
Text near the edges. Interfaces and crops cover it.
Trusting old size lists. Platforms change specifications.
Ignoring print resolution. A screen-sized image looks soft on paper.
Oversized files on websites. They slow pages without improving what visitors see.

Frequently Asked Questions

What size should AI-generated images be for social media?

Common starting points are 1080 by 1080 for squares, 1080 by 1350 for 4:5 portrait posts, 1080 by 1920 for vertical stories and 1200 by 630 for link previews. Platforms change their guidance, so confirm the current figures in each platform’s help pages and test on a phone before a campaign.

How do I get an AI image in a specific aspect ratio?

State the shape and the composition in the prompt, for example vertical 9:16 with the subject centred and empty space top and bottom. Then check the pixel dimensions of the file. If the ratio is slightly off, crop it. If it is wrong, regenerate with a clearer instruction.

Can I enlarge an AI image for printing?

Sometimes. Work out the pixels you need from the print size and dpi, for example about 2480 by 3508 for A4 at 300 dpi. If the image is smaller, upscale carefully and inspect for artefacts, or use it for layout drafts. For important print jobs, ask your printer for their requirements and bleed.

What is the best size for a link preview image?

About 1200 by 630 pixels, roughly 1.91:1, is the common choice for Open Graph images and works across most social networks and messengers. Keep text and faces away from the edges because some views crop, and keep the file small enough to load quickly, ideally a few hundred kilobytes.

Should I crop or stretch an image to fit a format?

Crop. Stretching changes the ratio by distorting the picture, which makes faces, logos and text look wrong. Generate with spare background around the subject so that you can crop to several formats, and export each format as its own file with a clear name.

The Bottom Line

AI image sizes are a planning problem, not a prompting trick: choose the ratio for each destination, ask for that shape with safe margins, check the real pixel dimensions and crop instead of stretching. Keep a short table of your channels and confirm current platform specifications before launch. You can build that table on the free plan, and see the Design plan for image generation on the Ask Mio pricing page.

AI Image Sizes and Aspect Ratios: A Platform Cheat Sheet