Compress Image

What Image Compression Can and Cannot Win

The only number that matters here is the one printed after the job finishes: source size, output size, and the change between them. This page is about making that number better, and about spotting the cases where it can't be improved at all.

Where the savings come from

A JPEG or WebP encoder throws away high-frequency detail the eye is bad at noticing (the noise in a blue sky, the subtle variation inside a shadow) and spends its bits on edges and flat regions instead. Photographs are full of that kind of detail, which is why a 6 MB camera file often drops to 800 KB with no visible change.

Screenshots, line art and logos have almost none. They're made of sharp boundaries and flat colour, which is exactly what lossy encoders handle worst, so as you lower the quality you get smeared halos around text long before you get a small file.

Where to put the slider

For photographs, the useful band is 75 to 85 percent. Below 60, the 8x8 blocks the encoder works in start to show, first around high-contrast edges. Above 90, the file grows quickly and you'll struggle to see any improvement. Encoding runs in a background worker, so a 24-megapixel photo won't lock up the tab while it's being processed.

PNG will disappoint you here

PNG compression is lossless. It looks for repeated pixel patterns and stores them once, and it never discards information, which is why the quality slider does nothing to a PNG. A photograph saved as PNG has very little repetition to exploit, so it stays large.

Switch the output format to WebP and the same picture typically falls by 60 to 80 percent. Only keep PNG when you need transparency without loss, or when the sharp edges of a diagram or screenshot have to be preserved exactly.

When the result comes back larger

This does happen, and it's not a fault. If the source was already compressed hard (anything downloaded from a social platform, for instance), a second pass at a higher quality than the original spends more bits than it saves. The tool steps the quality down and tries the encode again. If the output is still bigger, it tells you by how much and offers a button that gives you back the untouched original.

Take the original. Re-encoding a file that's already lossy piles new artefacts on top of the old ones and doesn't save you anything.

Metadata leaves with the pixels

Re-encoding rebuilds the file from decoded pixels, so camera make, lens, capture time and GPS coordinates don't survive. That's usually welcome. If you want to read those fields first, or remove them without re-compressing anything, Image Metadata Viewer does both and leaves the pixel data byte-identical.

Order of operations

If the image is going on a web page at 800 px wide, scale it down before you compress it. Fewer pixels saves more than any quality setting can, and the encoder then only works on the pixels that will actually ship. Resize Image handles that step.

If a PDF is heavy because of the photographs inside it, shrink those images and rebuild the document. Rasterising every page through Compress PDF would lose the selectable text along the way.