Resize Image

Exact dimensions are for when a container has a fixed size — a form field expecting precisely 600×800, a marketplace template, a print size in physical units converted to pixels. Aspect ratio stays locked by default, so setting the width computes a matching height rather than distorting the image; unlocking it stretches to the exact box, which is occasionally correct for a background fill and almost never correct for a photo of a person.

Percentage scaling is for batches of images that already have the right proportions and simply need to be smaller — a folder of photos of varying sizes that should all shrink by the same fraction rather than land at one fixed size. It treats every image proportionally regardless of its starting dimensions, which exact dimensions cannot do across a mixed batch.

Longest-edge mode caps whichever dimension is larger, which is the right choice for anything destined for web display, since layouts are almost always bounded by one dimension rather than both. It is also the only mode of the three that is safe to apply blindly to a batch of unknown, mixed orientations — portrait and landscape photos both get capped correctly without per-image adjustment.

Pixel count scales with the square of linear size, so an image resized to half its width and height has one quarter as many pixels — and file size tracks pixel count closely for both JPEG and PNG. This is a much larger reduction than adjusting a quality setting can achieve without visible loss, which is why resizing is usually the first thing to try when a file needs to be dramatically smaller rather than just a bit smaller.

It compounds with compression rather than replacing it. A 6000-pixel-wide camera photo resized to 1920px and then compressed at quality 80 will end up smaller than either step alone would produce, because the resize removes pixels the compressor would otherwise have to encode, and the compressor then removes redundancy within what remains.

Enlarging an image asks the resampling filter to invent pixel values that were never captured, and there is a hard ceiling on how well that can go: no algorithm recovers detail that the original resolution never recorded. A photo taken at 800×600 and stretched to 4K will fill the frame but will look soft rather than sharp, because the filter is interpolating between existing pixels rather than reconstructing lost ones.

That said, upscaling is not pointless. A modest enlargement — doubling a small icon, or scaling a low-res image up to meet a form's minimum dimension requirement — produces a result that looks perfectly acceptable at normal viewing distance even though it contains no new information. The failure mode is expecting a large upscale to look like a native high-resolution photo; it will not, and no tool can make it.