Compression is a trade, not a setting
Lossy compression works by discarding information the eye is poor at noticing. The quality value decides how much gets discarded, and there is no value that shrinks a file without giving anything up — only values where the loss is invisible and values where it is not.
The [msm_tool slug="image-compressor"] re-encodes the image at your chosen quality and reports the real output size, measured from the actual encoded file rather than estimated. That distinction matters because base64-length estimates are routinely off by a third.
Where the quality settings actually sit
- 90–100 — visually lossless and a large file. Worth it for photography portfolios and print sources, wasteful for a web page.
- 80–90 — where most photographs belong. Very hard to distinguish from the original at normal viewing size.
- 70–80 — visible artefacts on close inspection, usually acceptable for thumbnails and background images.
- Below 70 — visible artefacts in flat areas. Skies, skin and gradients show it first, because those are the regions with the least detail to hide the loss in.
The useful habit is to start at 85 and move down until you can see the damage, then back up one step — rather than picking a number and hoping.
Already-optimised images get larger
If an image has been compressed once at a given quality, re-encoding it at a similar quality produces a larger file, not a smaller one. The encoder is now trying to faithfully preserve the artefacts the first pass introduced, and artefacts are detailed, high-frequency information — exactly the kind that costs the most to store.
The tool detects this and says so rather than handing you a worse file with a smaller quality number attached. When you see that message, the answers are: accept the original, or go meaningfully lower on quality and accept visible loss.
Generation loss
Every lossy re-encode compounds. Compress, edit, compress again, and the damage accumulates in a way that is not visible at any single step and is obvious after four.
The rule that avoids it: keep an original in a lossless format or at maximum quality, do all editing from that, and compress once as the final step. Never edit a compressed file and re-save it as your new master.
Resize before you compress
This is the single largest saving available and it is regularly skipped.
A 4000-pixel photograph displayed in an 800-pixel column contains twenty-five times more pixels than anyone will see. Compressing it harder makes it blurrier at the same excessive size; resizing it to 1600 pixels — twice the display width, for high-density screens — cuts the file dramatically with no visible loss at all.
Resize first, compress second. In that order, quality 85 is usually plenty.
Format matters more than quality
Converting a JPEG to WebP at the same perceptual quality typically saves 25–35% on its own, before touching the quality setting. If you have both levers, use the format one first.
And check the format is right for the content: compressing a screenshot or a logo as JPEG produces a large file and soft edges, because JPEG is built for continuous tone. PNG or WebP is smaller and sharper for anything with flat colour and hard edges.
What compression cannot fix
A noisy photograph compresses badly, because noise is fine random detail and the encoder cannot tell it from real texture. Denoising before compressing helps considerably.
Likewise, an image that is already blurry does not get smaller for being compressed harder — the detail that would have cost bytes is not there to begin with.
A practical order
- Start from the highest-quality original you have.
- Resize to twice the display width.
- Choose the right format for the content — WebP for most web use.
- Compress once, at 80–85 for photographs.
- Look at the result at full size before shipping it.