What “without losing quality” really means
Honesty first: meaningful compression of photos is lossy — data is discarded. The trick is that "lossy" and "looks worse" are different things. Encoders discard detail below the threshold of human perception first, so there is a wide band of compression that is technically lossy and visually lossless: no difference you can find at normal viewing, even side by side.
The practical craft is staying inside that band. It is wider than most people think — typically 60–80% size reduction for photos.
The visually-lossless recipe
| Setting | Visually lossless zone | Notes |
|---|---|---|
| JPG quality | 80–85 | Below 75 differences become findable; below 60, obvious |
| WebP quality | 75–85 | WebP degrades more gracefully than JPG |
| Resizing | To the size actually displayed | The single biggest safe saving — see below |
| Chroma / color | Leave default (4:2:0) | Imperceptible in photos; encoders handle it |
Resizing is the underrated half. A 4000-pixel photo shown at 800 pixels was going to be downscaled by the browser anyway — doing it yourself first is pure saving with zero quality cost at the displayed size. Resize + quality 82 is how a 6 MB photo becomes 250 KB "without losing quality".
The four mistakes that cause visible loss
- 1. Recompressing exports. Each lossy save stacks loss on the last. Always compress from the original, never from a previous compression. (Why this happens.)
- 2. Quality settings below ~70 on photos with smooth gradients. Skies and skin show banding first.
- 3. JPG for text and sharp graphics. Halo artifacts around edges appear even at high quality — use PNG or WebP.
- 4. Upscaling. Enlarging an image cannot create detail; it creates soft mush that then compresses badly too.
A workflow you can trust
- Keep the original untouched — it is your quality ceiling forever.
- Decide the destination first — screen display? form upload? email? Each implies a size.
- One compression pass: resize to destination dimensions, quality 80–85, format per this guide.
- Check at 100% zoom if you're skeptical — flip between original and result.