How it works
Each file is opened by the browser as an ordinary image — the same decoder that shows pictures on web pages — so anything your browser can display can be read: JPEG, PNG, WebP, AVIF, GIF, BMP, ICO and SVG. A JPEG’s EXIF orientation is applied as it is decoded, so a portrait phone photo stays upright. The image is then drawn onto an off-screen canvas at its output size, with the browser’s high-quality smoothing, and the canvas is encoded with the format and quality you chose. Formats without transparency (JPEG) get the background colour painted under any transparent pixels first, instead of turning them black.
When the tool starts, it asks the browser to encode a one-pixel image in every format and checks what comes back, because a canvas asked for a format it cannot write silently returns a PNG instead. Formats that fail the test are disabled in the list rather than producing mislabelled files. Keep original writes each file in its own format where that is possible. Browsers cannot write GIF, BMP, ICO, SVG or HEIC: the first four become PNG, which keeps their transparency, and HEIC photos become JPEG.
Every result is compared with its input. When it is the same format at the same dimensions and not smaller, the original file is kept and labelled as such. Download all packs the results into a ZIP built in the page itself: the entries are stored rather than compressed again, since image formats are compressed already and a second pass would save almost nothing.