Batch Compress Images

Compression trades detail for size, and the useful question is never how much smaller a file can get but how much smaller it can get before anyone notices. That threshold depends far more on what the image contains than on the setting you choose.

Photographs compress well because lossy encoders are built for them: gradual tonal variation is exactly what the algorithm is designed to approximate cheaply. Flat graphics, screenshots, logos and anything containing text compress badly in the same formats, because the artefacts appear as halos around hard edges where the eye is most sensitive. A screenshot compressed as a photograph looks obviously wrong at a size where a photograph still looks perfect.

Quality settings are not a percentage of anything meaningful. A value of 80 in one encoder is not the same as 80 in another, and neither corresponds to keeping 80 percent of anything. They are positions on a curve, and the useful part of that curve is narrower than it looks: the difference between 95 and 85 is usually invisible and often halves the file, while the difference between 60 and 50 is frequently obvious and saves comparatively little.

Dimensions usually matter more than quality, and this is the lever people reach for last. An image displayed at 800 pixels wide and stored at 4000 pixels wide is carrying twenty-five times the pixel data the viewer will ever see. Resizing to the size it is actually displayed at typically saves more than any quality setting, and costs nothing visible at all because the extra pixels were never shown.

Compression is not reversible, so the rule is to keep originals and compress copies. A compressed file re-compressed is worse again, and there is no setting that recovers what the first pass discarded. For archival material, compress a derivative and leave the master alone.

The check that catches mistakes takes fifteen seconds: open the result and look at the part that matters most at full size, which for a product photograph is the texture and for a document scan is the smallest text. If it reads cleanly the compression was free; if it smudges, step back one setting.

Everything runs locally in your browser, which for image work means the files never leave the device and there is no upload time on a batch of hundreds.