AI Image Aspect Ratio & Size Guide for Every Platform (2026)

You generate a perfect square. Balanced, centred, lovely. Then you drop it into a Reel and the top of someone's head is gone, guillotined by a crop you never chose.

That's the most common mistake in AI imagery, and it happens after the good part is already done. The fix costs nothing and takes one decision: pick the aspect ratio before you generate, not after.

Here's why that habit matters more than any size chart. An AI model composes to the frame you give it. Ask for 16:9 and it arranges the shot for a wide screen. Ask for 9:16 and it builds for a phone held upright. Crop that wide shot into a vertical afterwards and you're throwing away pixels the model deliberately placed, then asking the leftovers to look intentional. They won't.

This guide hands you both halves in one place: the ratio to set in the prompt box, and the exact size it maps to on every platform that matters. One table, dated, built to be checked.

Why ratio matters before you prompt

AI generators don't paint one big picture and hand you a crop. They compose inside the exact rectangle you request. A portrait generation puts the subject where a portrait wants it, eyes high and headroom tight. A landscape generation spreads that same subject across a wide stage. The composition is baked into the shape.

Most models also hold their pixel budget roughly constant, near 1 megapixel, and trade width for height as the ratio changes. Midjourney's base render is about 1024×1024 at 1:1 and roughly 1456×816 at 16:9. Google's Gemini renders 16:9 at 1344×768. Same budget, redistributed. Push to an extreme ratio and that budget stretches thin along the long edge, which is why a 21:9 banner can look softer than a square from the same model.

Think of it like tailoring. You cut the cloth to the body from the start. You don't buy a tent and hack a suit out of it later. Cropping a square into a vertical is the tent approach: you get fabric where you need it, but the seams show.

So decide where the image will live, then generate at that shape. In JammyJar the shape picker sits right in the generate flow, so you set the destination before the model commits a single pixel.

Ratio, resolution and DPI in 90 seconds

Three words get muddled constantly, and every decision downstream depends on telling them apart.

Aspect ratio is shape: the proportion of width to height, written W:H. 16:9, 4:5, 1:1. It's scale-independent, so 1000×1500 and 2000×3000 are both 2:3.

Resolution is pixel count: the actual width times height, like 1080×1350. Two images can share a ratio and differ wildly in resolution. A 1:1 image might be 512×512 or 4096×4096, same shape, very different fitness for print or zoom.

DPI (dots per inch) only matters for print. It's the conversion between pixels and physical inches: pixels ÷ DPI = printed size. On a screen, DPI is a red herring. A 1200px-wide image displays identically whether its metadata says 72 or 300 DPI. The "72 DPI for web" rule you may have heard is folklore. Screens read pixels, full stop.

One number to keep: at the 300 DPI print standard, a 1024px image spans about 3.4 inches. That's the reality behind most AI output, and it decides whether you can print your image or only post it. More on that near the end.

Ratio is a creative choice. Resolution is a technical budget. DPI is a print-only sum. Keep them in separate boxes and the rest of this guide clicks into place.

The master size table

If you remember one number, remember 1080. Design your feed posts at 1080×1350 (4:5) and all your vertical video at 1080×1920 (9:16), and you're correct on most platforms most of the time. 1080px is the universal width baseline for 2026.

Now the full lookup. Every row lists the destination, its ratio, the pixel size, a note, and the ratio to set in your generator. That last column is the point of this whole guide.

When to reach for each shape:

• 1:1 square: the safe default for feed posts and product shots where the subject sits dead centre.
• 4:5 portrait: the reach play on Instagram, Facebook and LinkedIn, because it eats more screen than a square.
• 9:16 vertical: anything full-screen on a phone: Reels, Stories, TikTok, Shorts.
• 16:9 landscape: thumbnails, web heroes, in-stream X images, anything cinematic.
• 2:3 portrait: Pinterest's home turf, and the classic poster proportion.

One row above hides a genuine trap. A 3:4 profile-grid image publishes fine when you post it by hand, but Instagram's publishing API only accepts ratios between 4:5 and 1.91:1, so a scheduler rejects a 3:4 image outright. That one gets its own note below.

Ratio by platform

The table gives you the number. This section gives you the nuance the table can't hold.

Instagram: the 4:5 shift, the 3:4 grid, the scheduler trap

Instagram is the fussiest platform and the richest source of gotchas. Meta now favours 4:5 portrait (1080×1350) because it fills more of a phone screen than a square does. Adam Mosseri, who runs Instagram, announced native 3:4 support on Threads on 29 May 2025, as reported by 9to5Mac, framing 3:4 as the shape almost every phone camera shoots by default.

Here's the catch few guides mention. Instagram's publishing API accepts only ratios between 4:5 (0.8) and 1.91:1. A 3:4 image (0.75) sits just outside that window. Post it by hand and it appears fine. Hand it to a scheduler and the API rejects it. So upload 3:4 manually if you want the taller grid preview, but generate 4:5 for anything a tool will publish for you. Stories and Reels stay 9:16 (1080×1920), with anything important kept centre so the username bar up top and the reply field down the bottom don't cover it.

TikTok: 9:16, and mind the rail

TikTok is 9:16, 1080×1920, no argument. It'll accept 1:1 and 16:9, but it letterboxes them and they lose watch time. The catch is that TikTok publishes no official safe-zone template, so the working margins are community-tested: roughly 80–120px off the top, about 480px at the bottom for the caption and music credit, and around 110px on the right for the action rail. Keep faces, text and calls to action in the middle third and you'll clear all of it.

YouTube: two images, two rules

YouTube splits in two. The thumbnail is a fixed 1280×720 (16:9), under 2MB, minimum 640px wide, and it's judged at postage-stamp size in a feed, so design for legibility when it's barely 200px across. The banner is 2560×1440 on upload but crops hard across devices, leaving a guaranteed-visible safe zone of just 1546×423 in the centre. Some guides quote a 2048×1152 banner with a 1235×338 safe zone. The 2560×1440 figure is the better-supported one, so use it and keep the alternative in your back pocket.

Bonus info: picking a model that renders legible in-image text is a skill of its own. See our guide to the best AI tools for YouTube thumbnails.

Pinterest: taller wins

Pinterest rewards tall. The standard pin is 2:3 (1000×1500), and it's the only ratio Pinterest officially recommends for organic reach. Go taller than about 1:2.1 (1000×2100) and the feed crops you. Post a square and you take up roughly a third less space than a vertical pin. Pinterest also converts every upload to sRGB JPEG, so upload in sRGB for the cleanest result.

LinkedIn: its own pixel universe

LinkedIn lives by its own numbers. Its link-preview image is 1.91:1 (1200×627), a proportion it shares only with Facebook, while portrait 4:5 (1080×1350) now wins most mobile engagement. One deprecation to note: LinkedIn killed Stories on 30 September 2021, with product lead Liz Li conceding the team had assumed people wouldn't want informal video pinned to a professional profile. Any guide still listing LinkedIn Stories specs is stale. The company banner is contested, quoted as both 1128×191 and the older 4200×700. Use 1128×191 as the safer modern working size.

X and Facebook: the leftovers

X (Twitter) kept its header at 3:1 (1500×500) through the rebrand, with the profile photo overlapping the lower-left corner, so keep that area clear. In-stream images look best at 16:9 (1200×675). Facebook rides Meta's shared pipeline, so it mirrors Instagram: 4:5 in the feed, 9:16 for Stories and Reels, 1.91:1 for link previews. Its cover photo is the oddity, cropping to 851×315 on desktop but 640×360 on mobile, so keep the focal point dead centre. Meta doesn't publish official group or event cover sizes, so figures like 1640×856 are community working sizes, not gospel.

Web heroes and website images

On the web, two things matter: pixel dimensions and file weight. DPI doesn't come into it. Serve a full-width desktop hero at 1920×1080, step up to 2560×1440 only for 2× retina on wide containers, use 1600×900 for in-container banners, and make blog headers and Open Graph images 1200×630 so one asset covers both the page and the social preview.

Weight is where heroes quietly wreck your page speed. The hero is usually the Largest Contentful Paint element, so its file size is a ranking and experience issue, not just a nicety. Serve WebP, which Google's own WebP documentation puts at 25–34% smaller than a comparable JPEG at equivalent quality, or AVIF for smaller still. Target well under 400KB for a hero and under 200KB for inline images, and use srcset so a phone never downloads a 1920px file built for a desktop.

This is where format flexibility earns its keep. Generate the hero at 16:9, export it as WebP for the page, and export any logo or icon element as true SVG so it stays crisp at any size. JammyJar's Recraft Vector model outputs genuine editable SVG rather than a raster pretending to be one, which matters the moment you resize a logo and don't want it turning to mush.

Bonus info: choosing which model to run for which job is the whole game. Our best AI image generator guide ranks them.

Which AI models generate which sizes

Here's the part no social-media size chart will tell you: the models themselves cap out at wildly different resolutions, and that decides whether an image is print-ready or phone-only before you've typed a word.

They split into two rough tiers. A "1024px-locked" tier generates at about 1 megapixel and stops. A "flexible" tier lets you push to 2K or 4K natively. Get this wrong and you make a beautiful image that can never be printed larger than a postcard.

Those numbers come straight from vendor docs, and they move. Google's Developers Blog confirmed Gemini 2.5 Flash Image reached general availability on 2 October 2025 with 10 aspect ratios. Black Forest Labs shipped FLUX.2 on 25 November 2025 at up to 4 megapixels. ByteDance's Seed team lifted Seedream from 2K to 4K, up to 4096×4096, on 9 September 2025. Every release resets what "high-res" means.

The practical upshot: match the model to the job. Printable poster? Reach for a 4K-native model or Recraft V4 Pro. Drafting fifty thumbnail ideas fast and cheap? A 1024px model is fine. Logo you'll resize forever? Only true SVG will do. This is the call JammyJar's routing is built to make, so you're not memorising a resolution table every time you open the app.

Bonus info: the same subject phrased differently gets different results from each model. Our AI image prompting guide covers the dialects, and our consistent-style guide holds a look steady across a batch.

Ratio for AI-video follow-on

More and more, a still isn't the finished product. It's frame one of a clip. If a Reel or a Short is where the image is heading, the ratio decision moves a step earlier in the pipeline.

AI video models cluster around three shapes: 16:9 landscape, 9:16 vertical, and 1:1 square. Runway handles 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9, generating at 720p with an optional 4K upscale. Google's Veo does landscape and portrait with 4K output, though its API has had 9:16 quirks that quietly force some jobs back to 16:9.

The rule is the same one from the top of this guide, extended by a step. Generate the source image at the ratio the video will be. If the destination is a vertical Short, generate the still at 9:16 so the video model has a full-frame vertical composition to animate. Generate 16:9 and crop to 9:16 first, and you hand the model a stripped frame and force it to invent the edges it's missing. Decide the shape before you generate, then keep deciding it all the way down the pipeline.

Upscaling for print: the DPI maths

Here's the maths that decides whether your AI image can be printed at all: required pixels = print inches × DPI. At the 300 DPI print standard, a 1024px image prints at about 3.4 inches. Smaller than a postcard. Most AI generators output somewhere between 1024 and 2048px, so "gorgeous on screen, useless on paper" is the default state, not the exception.

Work the numbers backwards from the print size. A 20×30-inch poster at 300 DPI needs 6000×9000px. A 16×24-inch print needs 4800×7200px. An A4 sheet is 2480×3508px. Add about 35px of bleed per side (3mm at 300 DPI) so the cut never leaves a white sliver.

Two routes get you there. Generate natively large with a 4K-capable model (Gemini 3 Pro, Seedream 4.0, FLUX.2 or Recraft V4 Pro). Or generate small and upscale with an AI super-resolution tool, which reconstructs plausible detail rather than just stretching pixels.

Be realistic about upscaling. A working print shop, Little 6 Industries, puts the honest ceiling plainly: a 2× upscale is usually clean, 4× and beyond degrades visibly, and past roughly 12 inches from a 1024px source you should regenerate larger rather than upscale. Upscaler marketing tends to oversell how much real detail it can invent. Believe the print shop, not the ad.

For colour, consumer photo labs take sRGB. Only convert to CMYK when the print house asks, because CMYK shifts colour against RGB. Deliver TIFF or a high-quality JPEG, at 150–250 DPI for large-format work viewed at distance and 300 DPI for anything held close.

So before you commit to a poster, set the ratio and the resolution in JammyJar's shape picker up front, and pick a model that can actually reach the pixel count the printer needs.

Bonus info: our upscaling guide covers the generate-then-enlarge workflow, and our AI product photography guide handles catalogue-ready shots.

AI image aspect ratio FAQ

What's the best aspect ratio for AI images?
There isn't one. The best ratio is whatever your destination needs: 4:5 for most feed posts, 9:16 for anything full-screen on a phone, 16:9 for thumbnails and web heroes, 2:3 for Pinterest and posters. The point is to decide before you generate, because the model composes to the frame you request rather than cropping afterwards.

What's the difference between aspect ratio and resolution?
Aspect ratio is shape, the proportion of width to height, like 16:9 or 4:5. Resolution is the actual pixel count, like 1920×1080. Two images can share a ratio but differ hugely in resolution, and therefore in how they hold up when printed or zoomed. Ratio is a creative choice; resolution is a technical budget.

Does DPI matter for social media or web?
No. DPI only matters for print, where it converts pixels into physical inches (pixels ÷ DPI = printed size). Screens read pixel dimensions alone, so a 1200px image looks identical whether its metadata says 72 or 300 DPI. The old "72 DPI for web" rule is a metadata myth. Save DPI for the print shop.

Why does my AI image look low quality when I print it?
Because it's too small. Most AI models output around 1024px, which prints at just 3.4 inches at 300 DPI. To print larger, either generate with a 4K-capable model or upscale first. A 2× upscale is usually clean; 4× and beyond starts to degrade, so past about 12 inches you're better off regenerating at a higher resolution.

Can I just generate a square and crop it later?
For low-stakes reuse, yes. For anything that matters, no. Cropping discards pixels the model placed for the original frame and often decapitates subjects or breaks the composition. Generate per-ratio when composition counts, and crop a slightly oversized master only when you need one image across several shapes and quality is negotiable.

What size should I generate for Instagram?
Generate at 4:5 (1080×1350) for feed posts, which Meta now favours over squares, and 9:16 (1080×1920) for Stories and Reels. If you want the taller 3:4 grid preview, upload it by hand: Instagram's publishing API only accepts ratios between 4:5 and 1.91:1, so a scheduler will reject a 3:4 image.

Which AI model generates the highest resolution?
Among current models, Seedream 4.0 reaches up to 4096×4096, while Gemini 3 Pro Image, FLUX.2 and Recraft V4 Pro all hit around 4 megapixels or 4K. Lighter models like GPT Image Mini and standard Recraft outputs sit near 1024px, fine for screens but too small for large prints without upscaling.

Pick the shape first

Remember that square with its head cropped off in a Reel? That's not a resolution problem or a tool problem. It's a decision made in the wrong order.

Get the order right and everything downstream gets easier. Decide where the image will live. Set that shape in the generator. Generate at the model's native resolution. Upscale only if you're printing. Ratio is a creative choice, resolution a budget, DPI a print-only sum, and the whole mess behaves once you keep them straight.

So next time, before you type the prompt, pick the shape. Set the ratio in JammyJar's shape picker, generate for the destination, and let the crop tool gather dust.

Platform specs verified August 2026. These numbers move: check the current spec before anything you're publishing depends on it.