What the JPEG quality number actually means
Every export dialog has a quality slider from 1 to 100, and almost nobody knows what the number refers to. It is not a percentage of anything, not of the original, not of the file size, not of visual fidelity.
Understanding roughly what it does makes the difference between choosing a setting and guessing at one.
What JPEG actually does
JPEG compresses in four steps, and only one of them loses anything.
1. Convert to a brightness-and-colour representation. The image is converted from red-green-blue into one brightness channel and two colour channels. This alone loses nothing, but it enables the next step.
2. Optionally throw away colour resolution. Human vision is far more sensitive to brightness detail than colour detail, so JPEG can store the colour channels at half resolution. This is called chroma subsampling, usually written 4:2:0. It saves substantial space and is invisible on photographs, but it is very visible on sharp coloured edges, which is why text on a coloured background looks smeared in a JPEG.
3. Transform each 8x8 block into frequencies. Each small block is expressed as a set of patterns from smooth to fine. This is reversible and loses nothing.
4. Divide those frequencies by a table of numbers and round. This is the lossy step, and this is what the quality slider controls. It picks the table. Higher quality means smaller divisors, so less is rounded away. Fine detail is divided most aggressively, which is why JPEG artefacts appear around edges first.
So "quality 80" means "use the table associated with 80 in this encoder". Different encoders use different tables, which is why quality 80 in one program is not quality 80 in another.
Where the useful range is
For product photography on white:
- 95 to 100. Files two to three times larger with no visible benefit. Wasteful unless you are archiving.
- 85 to 92. The sensible working range. Visually indistinguishable from the original on product photographs.
- 75 to 85. Fine for supporting images. Artefacts start appearing around hard edges if you look closely.
- 60 to 75. Visible on close inspection. Acceptable when a file size limit forces it.
- Below 60. Artefacts visible, and, for white-background listings, the range where compression can push corner pixels off pure white. Combined with chroma subsampling on a subject near the frame edge, this is a real and measurable failure.
That last point is not theoretical. Testing a hard-edged subject 12 pixels from the frame edge, encoded at quality 60 with 4:2:0 subsampling, produced a corner reading (237, 255, 253), not white, and enough to fail an automated marketplace check. The same image at quality 82, or at quality 60 with subsampling off, passed.
Meeting a file size limit
When a marketplace caps file size, you have three levers, in order of how much they cost you:
- Lower quality. Costs fine detail first.
- Turn on chroma subsampling. Costs colour edge sharpness. Usually already on.
- Reduce dimensions. Costs everything, and usually the wrong trade for a listing where zoom matters.
A sensible tool walks quality down a step at a time until the file fits, and tells you where it landed. If your export settled at quality 55, you want to know that, because it means the image is near the range where corners stop being white.
Never re-save a JPEG
This is the practical rule that matters most.
Each save applies the lossy rounding again, to an image that already has rounding errors. Open a JPEG, crop it, save it, open it again, adjust brightness, save it again, and you have applied three generations of loss. The artefacts compound, particularly around edges.
Work from the original, RAW if you have it, otherwise the first JPEG the camera produced, and export once to the final file. If you must keep an intermediate, keep it as PNG or TIFF, which are lossless.
This is a real argument for batch processing over manual editing: a batch tool decodes once and encodes once, whereas a manual workflow through several programs can easily save four times without anyone noticing.
When not to use JPEG at all
Anything with transparency. JPEG has no alpha channel.
Flat graphics, logos, or text on solid colour. PNG will be both smaller and sharper, because JPEG's frequency approach is built for photographs and handles hard edges badly.
Anything you will edit again later. Keep a lossless master.
For photographs of products, JPEG at 85 to 92 remains the right answer almost every time, and newer formats help less than their file-size comparisons suggest once marketplace re-encoding is taken into account.