Step one

Compress

Drop in a photograph. It comes back as a JSON file named after your original, with the full accounting of where every byte went. Lossy uses the trained network and is scored against JPEG at a matched size. Lossless discards nothing and is scored against PNG.

Input

Lossy uses the trained network and reaches about 60x, by deciding what to discard. Lossless discards nothing and comes back bit for bit, at roughly 20 percent under PNG.

Lambda is the quality knob. Higher means a better picture and a bigger file.

Base85 fits about 6.1 bits per character, base64 fits 6. This is the cost of the container being text.

JPEG quality is binary searched until it lands on your file size, so neither codec gets a favourable operating point.

Your result will appear here.