How Much Can You Compress an Image Before It Shows?

Almost every guide lands on the same number: about 80. It is good advice and it is worth understanding why, because the number is not a percentage of quality, of file size, or of anything else. It is a position on a curve whose shape is what actually matters.

The curve, and why 80 sits where it does

Between 100 and roughly 85, file size falls steeply and almost nothing visible changes — the encoder is discarding detail the eye was never going to resolve. Between 85 and about 65 it keeps falling while damage accumulates slowly. Below 60 the size barely improves any further and the damage becomes obvious.

So the interesting region is narrow, and 80 sits inside it with room on both sides. Above that you are paying for invisibility; below it you are saving very little in exchange for something people can see.

What the damage actually looks like

Knowing the names makes the artefacts much easier to spot, and spotting them is the only way to judge a setting honestly.

• Blocking — the image dissolves into 8×8 squares, first visible in skies, walls and other large flat areas.
• Ringing — faint echoes hugging sharp edges, worst around text and thin lines against a plain background.
• Banding — smooth gradients break into visible steps, because the subtle values between them were the cheapest thing to discard.
• Colour bleed — saturated edges, particularly reds, smear slightly, because colour is stored at lower resolution than brightness.

That last one is not a defect but a design decision. Lossy formats store brightness at full resolution and colour at half or a quarter, because human vision is far more sensitive to the first. It is invisible on a photograph and immediately obvious on red text, which is one reason text should not be stored as a lossy image at all.

Different images have different limits

A busy photograph — foliage, crowds, texture — hides artefacts extremely well, because there is no smooth area for blocking to show against. The same setting applied to a portrait against a plain backdrop, or to a screenshot, produces something obviously degraded.

There is no correct quality setting, only a correct setting for this image. Judge the one in front of you.

Generation loss

Every lossy save discards a little more, and the artefacts of the previous save become detail the next one carefully preserves. Open, edit and re-save a JPEG a few times and the damage compounds visibly, which is the same effect as photocopying a photocopy.

Keep the original. Edit from it, and export a compressed copy each time rather than re-compressing yesterday's export. If an image has to go through several stages, keep it lossless until the last one.

Dimensions beat quality, usually by a lot

Before dragging quality down into the visible-damage region, check the size. Halving the width and height removes three quarters of the pixels, and it costs nothing at all if the image was never displayed larger than that. Compression can then stay in the comfortable range.

A 4000-pixel photograph at quality 60 will usually look worse and weigh more than the same photograph resized to 1600 at quality 85. Reach for the dimensions first.

How to judge it properly

Look at the result at the size it will be displayed, not at a thumbnail and not zoomed to 400 per cent. A shrunken preview hides exactly the artefacts you are trying to detect, and heavy magnification shows damage nobody will ever encounter. Then check the areas artefacts prefer: skies, shadows, skin, and any edge where dark meets light.