Guides
Image advice, measured rather than repeated
Most advice about image formats and compression is the same handful of rules of thumb copied between articles. We ran our own measurements instead — encoding a fixed set of images through the same code our tools use — and wrote up what the numbers actually showed, including the parts that contradicted what we expected. Measurements were last taken on 2026-08-09.
PNG vs JPEG vs WebP: the file sizes we actually measured
We encoded seven images to every format and measured the bytes. WebP beat JPEG by 12–70% depending on the picture — and JPEG lost badly on screenshots.
Read6 min readWhich JPEG or WebP quality setting should you actually use?
A full 10-to-100 quality sweep on seven images, with the measured file size at every step. The jump from 90 to 100 can triple the file for no visible gain.
Read7 min readResizing without losing quality: what actually happens to the pixels
We scored two downscaling methods against the mathematically exact answer. Lanczos won by 24–60% on photographs — and lost on flat graphics.
Read6 min readHow to shrink images for the web without them looking bad
Resize first, compress second, pick the format last. Following that order took our test photo from 2.2 MB to under 40 KB.
Read6 min readHow to make a white background transparent (and when not to)
Colour keying is perfect for logos and scans and useless on photographs. Here is how to pick a tolerance, and how to read a bad result.
Read6 min readHow in-browser image editing works, and why it is more private
Most "online" image tools upload your file to a server first. Here is what happens instead when the work is done in the browser — and how to check for yourself.
ReadHow we test
All measurements come from one script, scripts/benchmark.mjs, which runs in headless Chromium and encodes each test image through the browser’s own canvas encoder — the identical path our tools use, so the numbers reflect what you would actually get. Results are written to a JSON file that these pages render directly, so no figure on this site is typed by hand.
The test set is seven images: four public photographs from Wikimedia Commons, credited and linked in each guide, and three synthetic images generated by the script so that flat-graphic and screenshot content is exactly reproducible. The photographs are downloaded locally for measurement and are not redistributed here.