ResizeTo

Guides

Measured, not assumed

Most advice about image files is handed down rather than checked. Each of these started as a measurement on real photographs, and several of them ended up contradicting what we expected to find.

70 How the size search works Dimensions first, then quality by binary search, then more dimensions only if the ceiling still cannot be met. Why the floor sits at quality 70, and what happens when nothing fits. +27% Why re-saving a JPEG can make it bigger Five photographs, already compressed once, re-encoded at quality 92: every one grew by about 27%. Re-saving at the same quality ten times over cost nothing at all. q4 Fewer pixels, or lower quality? Forcing a 26-megapixel photograph under 200 KB on quality alone lands at quality 4. Removing pixels instead keeps quality 70. Why the two losses are not the same kind of loss. 22× JPEG, PNG, WebP or HEIC The same photograph came out 22 times heavier as PNG than as JPEG. WebP saved between 11% and 52%. What each format is genuinely for, and why forms still insist on JPEG. 41 What EXIF metadata reveals Forty-one hidden fields in one camera file: model, lens, editing software, operating system, timestamps, and on most phone photos the coordinates. Who strips it, who does not, and the one field you should not lose carelessly. 50 KB Photo and signature for an application form Why the two files are completely different compression problems, why cropping beats every compression setting, and the aspect-ratio trap that gets submissions rejected. 3× What size should images be on a website At the same width and the same quality, one photograph cost three times another. Measured weights at five widths, and why bytes fall more slowly than pixel counts.

How these were measured

Five photographs, chosen for range rather than convenience: a 26-megapixel camera original, a phone portrait, and three mid-sized files that had already been through a compression pipeline. Encoding is done by the browser's own encoder, the same one the tools on this site use, so the numbers describe what you get here rather than what a different library would produce.

Where a result contradicted the received wisdom, it is reported as measured. The generational-loss finding in the first guide is the clearest example: the popular account of it turns out to be wrong in the case that matters.