Image Compressor

Two levers, and only two

Size comes down in exactly two ways. The first is pixels: a 4000-pixel-wide photo shown at 1600 pixels on a website is carrying detail nobody will ever see, and scaling it down removes about 84% of the pixels with no visible change. The second is quality: JPEG and WebP spend their bits on the detail the eye is least sensitive to, so re-encoding at 75% instead of 95% typically halves the file while looking the same on a screen.

Both levers are on the form above, and the page reports what they did: the panel shows every picture with its size before and after, plus the pixel size that was written. Nothing hides behind a slider you cannot check.

Scaling uses area averaging rather than picking every Nth pixel: each new pixel is the average of the block it covers, which is what keeps downscaled photographs and lettering from turning jagged. Enlarging is never done, because interpolation cannot add detail that was never photographed.

Why "compressed" can sometimes come out larger

JPEG quality is a request, not a promise. A file that came out of a camera, a design tool or another compressor has already had detail removed, and re-encoding it finds less to remove than the encoder spends on headers and a second round of quantization. Measured here on a photo that had already been optimized for the web, a re-encode at 92% came out about 40% larger than the input.

That is why the tool always compares the file it made with the file you gave it, and why "hand my original file back" is the default: the download button then carries your own bytes, unchanged, and the result says so in plain words - "the re-encode came out at 900 KB, not smaller than the original 720 KB".

The same honesty applies in the other direction: the pictures that shrink a lot are the ones that were never optimized, like a photo straight from the camera roll or a PNG screenshot of a photograph. Expect anything from "no change at all" to "90% smaller", never a fixed number.

What re-encoding takes with it

A compressed picture is rebuilt from its pixels, so everything that was not pixels is gone: GPS coordinates, camera make and model, the timestamp, the color profile the phone recorded. For a picture headed for a website that is a privacy improvement - photos posted online have exposed home addresses through GPS tags - but it also means the change cannot be undone, so keep the originals if you may want them back.

The Exif orientation flag is handled rather than dropped: if a photo says "turn this 90 degrees", the pixels are turned so the picture is upright in any program, not only in viewers that honor the tag. When the same tag describes a mirror, the result says so, because a rotation cannot reproduce a mirror.

JPEG has no transparency channel at all, so a logo with a transparent background written as a JPEG gets the color you choose put behind it. Choose PNG if the transparency has to survive; the trade is a larger file, and the before-and-after numbers will show exactly how much larger.

Keeping, replacing or changing the format

The form can keep each picture in its own format, or force everything to JPEG, WebP or PNG. Keeping is the honest default - it changes the pixels without pretending the format was part of the problem; forcing WebP is where most of a modern saving lives, shown only where this browser can actually write it.

A format change comes with one catch the page names rather than hides: JPEG stores no transparency, so a see-through logo written as JPEG gets the background color you choose put behind it. The before and after numbers state exactly what the choice cost.