Divide 622 by 350 and you get 1.777142…
Divide 16 by 9 and you get 1.777777…
Those are different numbers. 622×350 is not 16:9. It is 311:175, which is a true statement, a correct reduction, and a completely useless thing to tell somebody.
This gap — between what a ratio mathematically is and what a person means by it — causes more confusion in image tools than any other single thing.
Why exact ratios almost never occur
An aspect ratio is a proportion. Pixel dimensions are whole numbers. Most ratios do not divide evenly into most numbers.
For 16:9 you need a width divisible by 16 and a height divisible by 9 in the same multiple — 1920×1080, 1280×720, 640×360. Crop by hand, or resize to fit a layout, or let a platform downscale your upload, and you land near one of those rather than on it.
Then a tool computes the greatest common divisor of 622 and 350, finds 2, and confidently reports 311:175. Which is arithmetically perfect and tells you nothing about whether your image is the shape you wanted.
What tools should do instead
Report the ratio the way a person would recognise it, and only fall back to arithmetic when there is nothing to recognise.
That is what our cropper does, in three stages:
- Name the common ratio if the dimensions are within 0.4% of it. 622×350 reports as 16:9, because for any practical purpose it is.
- Otherwise reduce to a fraction, but only if both terms stay small. 3:2, 5:4, 7:5 are meaningful. 311:175 is not.
- Otherwise give a decimal. “1.78:1” is honest and readable when nothing else fits.
This came out of a real bug. The readout originally did step two only, and a 622×350 crop — which is 16:9 to the eye and to any reasonable purpose — displayed as 311:175. Nothing was wrong with the crop. The tool was describing it in a way that made the user distrust a correct result.
The lesson generalises past image tools: a correct answer presented unrecognisably is a defect. Users cannot verify what they cannot read.
When the difference genuinely matters
Most of the time, being 0.03% off 16:9 is irrelevant. Three cases where it is not:
Video. Encoders and players expect exact dimensions, and many require even numbers for chroma subsampling. Odd-numbered dimensions are rejected outright by some pipelines. Video wants exact.
Platform specifications. A service that says 1200×630 means those numbers. Supply 1198×629 and it will be scaled or cropped by an algorithm you do not control, usually from the centre, usually at the moment you would rather it did not.
Repeated operations. One rounding is negligible. Crop, resize, crop again and the roundings accumulate. This is the same compounding problem as re-saving a JPEG, applied to geometry instead of colour.
Fitting a box, and the thing people get wrong
Give a resizer both a width and a height and there are two possible interpretations.
Fit inside — scale until the image fits entirely within the box, preserving the ratio. You get the whole image, with the box unfilled in one dimension.
Fill and crop — scale until the image covers the box, preserving the ratio, then trim the overflow. You fill the box exactly and lose part of the picture.
Stretch — ignore the ratio and force both dimensions. The image distorts, and everybody notices even if they cannot say why.
Our resizer fits inside when given both dimensions, and links the two by default so changing one updates the other. Fit-inside is the safe default: it never silently discards part of your image, and never distorts it. If you need the box filled exactly, crop first to the target ratio, then resize — two deliberate steps rather than one ambiguous one.
Crop from the original, not the preview
A subtle one that separates tools that work from tools that mostly work.
A cropper shows you a preview, which is usually a scaled-down version of your image — a 4,000-pixel photograph displayed at 800. You drag a box on that preview.
If the tool takes the crop from the preview, your output is 4,000-pixel-quality dimensions rendered from 800 pixels of data, upscaled and soft. If it maps your selection back onto the full-resolution original and crops from there, you get every pixel the source had.
Ours does the latter. It is worth verifying in any cropper you rely on: crop a small region of a large image and check the output dimensions and sharpness against expectation. A tool cropping from the preview gives itself away immediately.
The ratios worth knowing
| Ratio | Decimal | Where it turns up |
|---|---|---|
| 1:1 | 1.000 | Profile pictures, product grids |
| 4:5 | 0.800 | The tallest most feeds will show |
| 3:2 | 1.500 | Most stills cameras |
| 16:9 | 1.778 | Video, screens, most link previews |
| 1.91:1 | 1.910 | Common social link-preview crop |
The last is not a clean fraction at all, which is the whole point of this article in one line: real-world ratios are frequently decimals, and a tool that insists on fractions will report them unrecognisably.
Portrait, landscape, and the ratio that inverts
One source of confusion worth clearing up: ratios are conventionally written width-first, so 4:5 is portrait and 5:4 is landscape. They are different shapes, not two ways of saying the same thing.
Tools disagree about this. Some present a ratio and rotate it to match your image’s current orientation; others hold it fixed and expect you to pick the right one. Neither is wrong, but if a crop comes out ninety degrees from what you intended, this is usually why.
The safe habit is to check the resulting pixel dimensions rather than trusting the label. A readout saying “16:9” is describing a shape; the numbers underneath tell you which way up it is.
There is a related trap in the “original ratio” option most croppers offer. It locks to the ratio of the file you loaded — which, if that file was already cropped by a platform or resized by an editor, is not the ratio you think it is. Locking to “original” preserves whatever shape the image currently has, including a shape you did not choose.
Rounding, and which direction to round
When a target ratio does not divide evenly into your dimensions, you have to round, and the direction is not arbitrary.
Round down, not up. Cropping 622×350 to a true 16:9 means going to 622×349.875, which is not a thing — so you go to 620×349 or accept the 0.03% error. Rounding up means inventing pixels that were not in the source.
Adjust the dimension you care about less. If the width is fixed by a layout, adjust the height. If a subject must not be cut, adjust the other axis.
Keep dimensions even where you can. Some encoders, particularly video, require it, and some image pipelines handle even dimensions more predictably.
Do the arithmetic once. Crop to the ratio and resize to the size in one considered step rather than three approximate ones, for the same reason you should not re-save a JPEG repeatedly — each rounding is small, and they accumulate.
The dimensions worth having to hand
Platform specifications change, so verify before anything important, but these are the shapes that recur.
| Use | Typical dimensions | Ratio |
|---|---|---|
| Link preview image | 1200×630 | 1.91:1 |
| Square post | 1080×1080 | 1:1 |
| Portrait post | 1080×1350 | 4:5 |
| Video, 1080p | 1920×1080 | 16:9 |
| Video, 4K | 3840×2160 | 16:9 |
Notice that all five are exact — the platforms specify dimensions that divide cleanly, precisely so their pipelines do not have to round. Which means that when your image is not one of these, the platform will resize it, and the resize will be its choice rather than yours.
Supplying the exact numbers is the only way to keep control of the crop. It is also the reason a tool that reports 622×350 as “16:9” is being helpful rather than sloppy: it is telling you the shape is right, while the exact dimensions still need attention if a platform is going to be strict about them.
Crop to a locked ratio on a live preview, taken from the full-resolution original, with our image cropper. Nothing is uploaded.