Do these in order
The order matters more than any individual technique, because each step shrinks the work the next one has to do.
Most people go straight to the quality slider, which is the smallest lever with the most visible damage.
1. Resize — usually the biggest win
File size scales roughly with pixel count. Halve both dimensions and you remove about three quarters of the data, before any compression decision has been made.
A photograph straight from a phone is commonly 4000 pixels wide. Displayed in a 600-pixel column on a web page, more than 90% of those pixels are being downloaded and then thrown away by the browser.
The rule: match the largest size the image is actually displayed at, then double it for high-density screens. A 600-pixel column wants a 1200-pixel image.
Resize the image first. Everything after this is working on a much smaller starting point.
For print, ignore this step — printing genuinely needs the pixels.
2. Change the format
This is the second-largest lever, and it costs nothing visually.
Measured on this site’s fixed test image, from the same source:
| Format | Size |
|---|---|
| PNG | 3.97 MB |
| JPG | 375 KB |
| WebP | 247 KB |
| AVIF | 109 KB |
Converting a photographic PNG to WebP took 94% off it. That is not a saving any quality slider will reach without visible damage.
- Going on a website you control → WebP or AVIF
- Being sent to a person → JPG, because you cannot predict their software
- A screenshot with text → keep PNG and optimise it losslessly
3. Now lower the quality
Only after the first two steps, because they have already done the heavy lifting.
Compress the JPG and watch the size readout after each change. Two things that page’s measurements make clear:
- Re-encoding at the quality a file already uses gains nothing. Our test file was saved at quality 82; re-encoding at 82 produced the same size.
- Raising quality above the source’s own level makes the file bigger without improving it. At 95, the same file grew 54% — spending bits faithfully recording artefacts that were already there.
So you have to go below whatever the source used, and you usually cannot tell what that was. Which is why the readout matters more than the number on the slider.
4. Check it at full size
Compression damage is invisible in a thumbnail and obvious at 100%.
Look at two things specifically: flat areas such as skies or plain walls, where banding and blotching appear first, and hard edges, where haloing shows. If either looks wrong, go back one step rather than accepting it — the damage is permanent.
Common situations
Email attachment too large → resize to 2000 pixels wide, convert to JPG at quality 82. That handles almost every case.
Web page loading slowly → resize to twice the displayed width, convert to WebP or AVIF, serve both with a fallback.
Upload form rejecting the file → resize first, then lower quality while watching the readout until it fits.
Screenshot too big → optimise the PNG losslessly. Keeps the text perfectly crisp. Convert to WebP only if slight softening is acceptable.
iPhone photos too large → convert HEIC to JPG, resizing at the same time if they are headed for a screen.
What not to do
Do not convert to PNG to “keep the quality”. For a photograph this makes the file roughly ten times larger and improves nothing — PNG faithfully preserves the detail loss that is already there.
Do not compress repeatedly. Each lossy save discards more. Go back to the best original you have and do it once.
Do not resize upward. Enlarging invents pixels by interpolation; it cannot recover detail that was never captured.