JPG compression: how far to push it before quality drops

JPEG’s quality slider runs from 0 to 100, and it’s tempting to assume the relationship between that number and file size is a straight line — drop from 90 to 45 and expect roughly half the bytes. It isn’t linear. Most of the possible file-size savings happen in the top third of the scale, between 100 and about 75, because JPEG’s encoder is aggressively removing redundant high-frequency detail that the human eye barely registers. Below 75, further quality drops buy diminishing size savings while visible artifacts start climbing fast.

How JPEG compression actually works

JPEG (defined in the ITU-T T.81 standard) splits an image into 8×8 pixel blocks, runs each through a discrete cosine transform, and then quantizesthe result — dividing each frequency coefficient by a number from a quantization table and rounding. The quality setting scales that table: higher quality means finer quantization (less rounding, more preserved detail), lower quality means coarser quantization (more rounding, more discarded detail, smaller file). The 8×8 block structure is also why heavy compression produces visible square “blockiness” rather than a uniform blur — you’re seeing the block boundaries where quantization error doesn’t align smoothly with its neighbors.

Where artifacts actually start showing up

Content matters as much as the number. A smooth photo of a sunset sky shows banding artifacts much earlier than a busy, high-detail photo of foliage, because JPEG’s quantization error is most visible in large flat color regions. Screenshots with sharp text edges also degrade earlier than photos — this is one reason PNG, not JPEG, is the right format for screenshots and UI mockups.

Quality vs. scale: two different levers

Quality and pixel dimensions are independent controls that both shrink file size, and conflating them is a common mistake. Dropping the quality slider keeps every pixel but discards fine detail within each one; scaling the image down (say, to 50% of its original width and height) removes pixels entirely. For a photo that will only ever display at 800px wide, scaling to 800px first and then compressing at a higher quality setting almost always looks better — and produces a smaller file — than keeping the full resolution and crushing the quality slider to compensate.

Testing it yourself

The only reliable way to pick a setting for a specific image is to compare outputs directly, since the “safe” threshold shifts with content, viewing size, and screen. An image compressorthat shows before/after file size and lets you toggle quality while previewing the result removes the guesswork — start around 80, and back it down only if the size still isn’t low enough for your use case. For anything with sharp edges, transparency, or text, check whether a different format would serve you better before tuning JPEG quality further.