Quality or target size?
Quality is for “lighter, but still good”: 80% (the default) is hard to tell from the original, 60% gives much smaller files for a web page or a chat. Target size is for forms that set a cap (an ID photo under 200 KB, an attachment under 1 MB): type the size and each image is compressed to land just under it, with the highest quality that fits.
How each format is compressed
JPG and WebP are re-encoded at a lower quality. PNG is lossless, so it is reduced to at most 256 colours instead (the technique of pngquant and TinyPNG), transparency kept: screenshots and logos often lose 50–80% with no visible change. Each image keeps its format; animated WebP and PNG files are left untouched, and an image that would not get smaller is kept as it was.
Dimensions, metadata and limits
Compression doesn't change the pixel dimensions: a 4000 px photo stays 4000 px. For a website, resizing it to the width it is displayed at saves even more, so use Resize Image first. Metadata such as GPS position and camera model is removed, the orientation is kept. Up to 50 MB per image.
Understanding image compression
Image compression reduces the amount of data needed to store a picture. There are two approaches: lossy compression discards visual information the eye rarely notices, while lossless compression stores the same pixels more efficiently without removing anything. JPG is lossy, WebP can be either, and PNG is lossless — which is why this tool lowers the quality of JPG and WebP files but reduces the number of colours of a PNG instead.
When you compress a JPG image, the algorithm divides the picture into 8×8 pixel blocks and applies a mathematical transformation (DCT) that separates coarse colour information from fine detail. By discarding high-frequency detail — subtle texture and noise the eye glosses over — the encoder produces a dramatically smaller file. The quality slider controls how aggressively this detail is removed: at 90% quality barely any loss is perceptible, while at 55% you may notice slight softening around sharp edges.