PNG

What it is

PNG compresses in two stages. First it filters each row, storing the difference from the pixel to the left or above rather than the absolute value — in a smooth gradient those differences are nearly all the same small number. Then it runs DEFLATE over the result, which finds repeated byte sequences and replaces them with back-references.

Decompression reverses both exactly. A PNG is bit-for-bit identical to what went in, which is what lossless means. It also carries a full 8-bit alpha channel, so transparency is graduated rather than a single on/off colour.

Why it is spectacular on some images and terrible on others

PNG exploits repetition, and its performance is entirely a function of how much repetition the image contains. A screenshot is thousands of identical background pixels, the same anti-aliased letterform recurring, flat panels of interface colour — enormous redundancy, so a 1920 x 1080 screenshot might be 180 KB.

A photograph of foliage has almost none. Sensor noise alone means no two identical-looking pixels are actually identical, so there is nothing to reference. The same 1920 x 1080 frame can be over 4 MB as PNG and around 290 KB as a quality-80 JPEG — fourteen times smaller, and visually the same.

What to use it for

Screenshots, diagrams, charts, line art, logos, UI assets, anything with text, and any image where transparency has to stay crisp. Also masters and archives, where the point is that nothing is discarded.

Not photographs for delivery. A common error is treating PNG as the high-quality option and JPEG as the low-quality one; they are matched to content, not ranked. Losslessly converting a quality-60 JPEG to PNG faithfully preserves quality-60 artefacts in a much larger file.

When lossless WebP is better

For the same content, on the web, lossless WebP is typically 20-30% smaller than PNG with identical pixels and the same alpha support. The reason to still choose PNG is compatibility: it is accepted by essentially everything, and WebP is not.