Lossy compression

Deliberately discards information the eye perceives poorly.

Uses psychovisual models to reduce data the human eye barely registers. JPG, WebP-Lossy and AVIF work this way. Invisible at the right quality setting (78–82 for JPG).