WebP timing turned out to be true on a few years old tablet, but not on a newer phone nor desktop; the number of large cores determines that time for JXL. The CPU time is probably lower for WebP, didn't measure that. But wall clock tells a different story, and the encode times and size bury WebP.
AVIF is a strange format since its lossy compression can _improve_ when you add a smaller layer; it's smaller than WEBP in the normal modes (s6; s>=7 switches to real-time mode which sucks and s3 takes centuries), sure. But even it decodes faster than WebP in favorable chip conditions. And JXL via Jixel just eats both for breakfast with plenty of room to spare.
WebP timing turned out to be true on a few years old tablet, but not on a newer phone nor desktop; the number of large cores determines that time for JXL. The CPU time is probably lower for WebP, didn't measure that. But wall clock tells a different story, and the encode times and size bury WebP.
AVIF is a strange format since its lossy compression can _improve_ when you add a smaller layer; it's smaller than WEBP in the normal modes (s6; s>=7 switches to real-time mode which sucks and s3 takes centuries), sure. But even it decodes faster than WebP in favorable chip conditions. And JXL via Jixel just eats both for breakfast with plenty of room to spare.