How to resize images correctly

Two different operations get called “making the image smaller”, and confusing them is the source of most resizing complaints.

• Resizing scales the whole image. Everything stays in frame; the pixels are rebuilt at a new size.
• Cropping cuts pixels away at the edges. What remains is untouched, but you have lost part of the picture.

If you need a specific shape — a square avatar from a landscape photo, say — you need a crop, because resizing alone cannot change the proportions without distortion. Use the cropper to choose the region, then the resizer to set the final dimensions. Aspect ratios in more depth.

A 3:2 photograph forced into a 1:1 box is stretched — faces widen, circles become ovals, and everyone can tell even when they cannot say why. Keeping the ratio locked is the default for a reason.

The practical habit: set one dimension and leave the other blank. Give a width of 1600 and the height follows from the source proportions. That way a folder of mixed portrait and landscape images comes out consistently sized on the constrained edge rather than all squashed into one box.

Downscaling discards information, which sounds bad and is actually the easy direction — the result is sharp and clean as long as the sampling is done properly.

• Go in one step, from the original. Repeatedly halving through several saved intermediates compounds softness and, in a lossy format, compression artefacts too.
• Downscale before compressing, not after. Compressing first bakes in artefacts that then get scaled and blurred.
• Expect to sharpen a little. Any good downscale softens fine detail slightly; a small amount of output sharpening restores the bite.

The resizer here scales in high quality via the browser canvas, which is fine for the sizes web and stock delivery need. Files never leave your device, and a whole batch takes the same settings so a set stays consistent.

Enlarging invents pixels between the ones you have. It cannot add detail that was not recorded, so the honest expectation is a bigger, softer image rather than a better one.

• Up to about 20% is usually unnoticed, especially on photographic content with little fine texture.
• Doubling is visible on hair, fabric, foliage and any text in frame.
• Upscaling to satisfy a stock agency's minimum resolution tends to be rejected — interpolated detail has a recognisable look.

If you need more pixels than you have, the reliable answers are to reshoot, print smaller, or accept a lower density. Working out what you need.

Doing the same steps in the wrong sequence costs quality for nothing. The order that works:

1. Crop to the region and ratio you want, from the original file.
2. Resize to the final pixel dimensions.
3. Set DPI, if the destination is print. Why it is last.
4. Compress or convert format, once, as the final export.

Every step after a lossy save is a step applied to damaged pixels, which is why compression sits at the end rather than anywhere else.

If the constraint is a file size rather than a dimension, resize generously and let the KB/MB reducer find the quality — it downscales further only when it has to.