How to Choose a JPEG Quality Setting

The number is not a percentage

Quality 50 does not mean the image keeps half its quality. The number indexes into a quantisation table that controls how coarsely each frequency component is rounded off.

Different encoders use different tables, so quality 80 in one application does not match quality 80 in another. The numbers are roughly comparable in direction but not in value.

This is why advice phrased as a specific number is always approximate, and why the useful method is to look at the resulting file and its size rather than trusting the setting to mean something absolute.

Quality 100 is not lossless

This surprises people regularly. At quality 100 the quantisation is minimal but still present, and the colour subsampling step may still be discarding chroma detail depending on the encoder.

What you get is a very large file that is still lossy. A photo at quality 100 can be three times the size of the same photo at quality 90 with no difference any person could identify in a blind comparison.

If you genuinely need lossless, JPEG is the wrong format — use PNG or lossless WebP. If you need JPEG, 95 is the sensible ceiling and everything above it is paying for nothing.

Settings by destination

These are starting points, not rules, and they assume a reasonably standard encoder.

• Archive or master copy: do not use JPEG at all. Keep the original.
• Print: 90 to 95. Artefacts that vanish on screen can appear on paper under close viewing.
• Portfolio or photography site where the work is the point: 85.
• General web content, blog images, product photos: 80. This is the workhorse setting.
• Email attachments and messaging: 70. The recipient is viewing small.
• Thumbnails and previews: 60. Nobody inspects a thumbnail.
• Hard KB target for a form: whatever hits the number, but resize first.

Content changes the answer more than destination does

Two photos at the same setting can look completely different in terms of visible damage. Smooth gradients — skies, studio backdrops, out-of-focus backgrounds — show banding first, and often need 85 where a busy photo is fine at 70.

Busy, detailed content hides artefacts well, because there is so much going on that the eye has nothing clean to compare against. Foliage, crowds, and textured surfaces tolerate aggressive settings.

Anything with text or hard graphic edges is the worst case. JPEG puts ringing around every edge, and no quality setting below about 90 hides it. For those, the right answer is usually a different format entirely.

Where the curve bends

File size against quality is not linear. Going from 95 to 85 typically removes half the file with essentially no visible change. Going from 85 to 75 removes another third with very little. Going from 75 to 65 starts to show on smooth areas.

Below about 50, artefacts are obvious to anyone looking: blocking in flat regions, mosquito noise around edges, and colour banding in gradients.

The practical consequence is that 80 to 85 is where most of the free saving lives. Pushing below 70 buys progressively less size for progressively more visible damage.

Chroma subsampling matters more than people realise

JPEG usually stores colour information at half resolution horizontally and vertically, because human vision is far more sensitive to brightness than to colour. That is called 4:2:0 subsampling and it is a large part of why JPEG is efficient.

It is invisible on photographs. It is very visible on saturated coloured text or graphics, where a red heading on a white background develops fringing that looks like a printing misregistration.

Some encoders let you disable it, usually labelled 4:4:4 or best quality colour. If you are compressing something with coloured text and it looks wrong at high quality settings, subsampling is the likely culprit rather than the quality number.

Test on the worst image, not the best

If you are choosing a setting for a batch, pick the hardest image in it — the one with the most sky, the smoothest gradient, or the most graphic content — and set quality by that.

Judging at full size matters. Artefacts invisible in a thumbnail are obvious at 100 percent, and the reverse trap also exists: an image that looks damaged when zoomed in may be perfectly fine at its display size.

Look at the actual context. A photo in a 600-pixel column tolerates far more compression than a full-bleed hero, and setting one policy for both wastes bytes on one and quality on the other.

Compress once

Every re-save compounds. Editing a JPEG and saving it again re-quantises an already-quantised image, and the damage accumulates in a way that is not reversible and not visible until it is bad.

Do all editing on a lossless master and export to JPEG once per destination. If the master is gone, minimise further saves rather than trying to repair.

The one exception worth knowing: lossless JPEG operations like rotation by 90 degrees and cropping on block boundaries exist in some tools and do not re-encode. Ordinary edit-and-save does.

Frequently asked questions