2026-09-15
An uncompressed photo is enormous. A 12 megapixel image needs roughly 36 MB before you do anything to it, which was unusable on 1990s connections and is still wasteful today.
JPEG was built on one insight: human vision is far more sensitive to brightness than to colour. So it keeps brightness detail and throws most colour detail away. That single trade-off does most of the size reduction before any quality slider is involved.
Contrary to how it is marketed, the quality setting does not change how the algorithm works. It changes the size of the quantisation step. Higher quality means finer frequency detail is preserved; lower quality rounds it away.
This is why quality reductions are not linear. Dropping from 100 to 90 costs almost nothing visually. Dropping from 50 to 40 is where visible damage usually appears.
Sharp edges contain high-frequency energy. Quantisation removes high frequencies first, so the parts of the image with the most high-frequency content lose the most. That is exactly where the familiar blocky and ringing artefacts show up.
It also explains why text and line art compress badly. They are almost entirely high-frequency content, which is the first thing JPEG discards.
Never edit and re-save a JPEG repeatedly. Every save applies quantisation again to already-quantised data, and the damage accumulates. Keep a lossless master and export a JPEG only at the end.
No. JPEG is lossy at every setting, including 100. For truly lossless storage use PNG or a lossless format.
Not exactly, but the loss does compound. The second save quantises data that already lost detail, so the visible decline is often larger than the first save.
For flat-colour graphics and screenshots, PNG can win, because JPEG is spending bits trying to represent sharp edges that PNG stores exactly.