Workflow
How to compress an image — and what to check before you send it
Compression is not one setting. It is four decisions in a fixed order, and doing them out of order is why files end up either tiny and ugly or sharp and rejected.
At a glance
- Decide the destination first: pixel dimensions come from where the image will be shown, not from the file you happen to have.
- Resize before you compress. Dropping quality to reach a size target blurs a full-size image instead of removing pixels nobody sees.
- For photographs, quality 75-85 is the useful band; above 90 the file grows several times over for changes the eye cannot detect.
- Never compress an already-compressed file twice. Export once from the largest original you have.
Start with the destination, not the file
Two questions decide everything that follows: how many pixels will the viewer actually see, and what limit does the destination enforce? A profile picture shown at 200 x 200 pixels does not need 4000 pixels of width; a poster printed at A3 does. A job application portal that rejects anything over 1 MB is a hard constraint; a photography portfolio is judged by eye instead.
Write both down before touching a slider. Most failed compressions are a mismatch between those two facts — someone tries to hit a byte limit by destroying detail that the destination would have displayed anyway, or someone compresses so gently that the file still does not fit.
Resize first, always
Compression removes information from every pixel. Resizing removes pixels that were never going to be shown. When you have to choose between them, resizing wins by a wide margin, because it reduces the raw data the encoder has to describe.
Take a 4000 x 3000 phone photo that needs to fit a 2 MB limit. Keeping it at full size forces the encoder into a low quality band where blocking becomes visible on every edge. Downsampling to 1600 x 1200 first removes 84% of the pixel data, and the quality slider can then sit at a comfortable 80 with room to spare. Same subject, same screen size in practice, far better result.
Crop before you resize when the aspect ratio has to change. Cropping first means every pixel you keep is a pixel you want; resizing first and cropping afterwards throws away work and can soften the crop edge.
Then set quality — in the band where it still matters
The quality slider controls how much detail the encoder is allowed to discard. Its relationship to file size is steeply non-linear, and it saturates earlier than most people assume. On a 1600 x 1200 photo-like test image, moving from quality 75 to 90 multiplied the file by 2.94 while the structural similarity score moved from 0.888 to 0.893.
| Setting | Typical use | What it costs |
|---|---|---|
| q60-q70 | Thumbnails, previews, chat stickers | Visible softening on fine texture and hair |
| q75-q80 | Web images, feed posts, documents | The practical sweet spot for photographs |
| q82-q88 | Portfolio and print proofs | Roughly twice the bytes of q75 for a small gain |
| q90+ | Archival or intermediate editing files | Multiples of the size, and no better on screen |
One caution specific to JPEG: quality above roughly 90 stops being a quality decision and becomes an archival one. On our test photo the measured PSNR actually fell at q95 (38.0 dB) compared with q85 (38.9 dB) — the encoder was faithfully preserving grain and noise, not detail.
Choose the format for the content, not by habit
JPEG is the right default for photographs. It is a poor choice for anything with text, thin lines or flat colour. Screenshots make that plain: from q40 to q95 our 1920 x 1080 UI capture grew 170.9 KB to 752 KB while PSNR stayed flat at 26.7 dB — four times the bytes for a metric that did not move, because JPEG spends its budget approximating the high-contrast edges of glyphs.
For text-heavy images, take a lossless or near-lossless path instead: PNG for flat graphics and text where exactness matters, or WebP in its lossless mode when the destination accepts it. For flat illustrations, lossless formats are also competitive with JPEG because large areas of identical colour cost almost nothing to describe.
Verify before you send
The check that catches almost every mistake takes ten seconds:
- Open the exported file in a viewer at 100% zoom, not fitted to the window. Fit-to-window hides exactly the artefacts that compression creates.
- Look at the edges: text, straight lines, and the boundaries between very different colours are where blocking and ringing appear first.
- Confirm the numbers: pixel dimensions, byte size, and the file extension the destination expects.
- If a destination strips or rejects formats, send the least surprising one — JPEG for photographs, PNG for graphics.
Mistakes that cost people their upload
Compressing twice. Exporting a JPEG from a JPEG throws away information that was already marginal. Work from the largest original you have and export once.
Oversharpening after compression. Sharpening is high-frequency detail, which is exactly what the encoder discarded. Applying it afterwards amplifies the artefacts instead of restoring the detail.
Chasing a byte target with quality alone. If the target is far below what the dimensions allow, no quality setting gets there cleanly. Resize first; then, if it still misses, reduce dimensions again.
Trusting an online compressor that re-encodes. Some tools recompress at a fixed internal setting and discard your file's other properties. Working locally means you can see the same file you are about to send.
Doing it in the browser
The image compressor on this site resizes and compresses in one pass, on your device: choose the output dimensions, pick a format, watch the reported output size as you move the quality slider, and compare the result against the original before downloading. Nothing is uploaded, so there is no queue, no account, and no copy of your file on someone else's disk.
Resize, compress and convert images in your browser
Everything happens on your device — no upload, no account.