[just_for_you]

To add to this, it's already possible to give this a try via building Brunsli from source:

https://github.com/google/brunsli

I personally use the cbrunsli/dbrunsli cmdline programs to archive old high-resolution JPEG photos that I've taken over the years. Having a gander at one subdirectory with 94 photos @ 354 MB in size, running cbrunsli on them brings the size down to 282 MB, which brings in savings of about 20%. And if I ever wanted to convert them back to JPEG, each file would be bit-identical to the originals.

Perhaps it's a little early to trust my data to JPEG XL/Brunsli, but I've ran tests comparing hundreds of MD5 checksums of JPEG files losslessly recreated by Brunsli, and have not yet ran into a single mismatch.

I can only say that I am very excited for the day that JPEG XL truly hits primetime.

[JyrkiAlakuijala]

Brunsli works very well, but is not compatible with the final JPEG XL format. For being able to reduce the binary size of libjxl we decided to use a more holistic approach where the VarDCT-machinery is used to encode jpegs losslessly. This saved about 20 % of the binary size and reduced attack surface. Now the decoder binary size is about 300 kB on Arm.