Considerations for adding cjpegli as potential jpeg encoder?

[Earnestly]

The team who made jpegxl's cjxl have incorporated much of their discoveries into a tool called cjpegli for use with jpeg. https://github.com/libjxl/libjxl/blob/main/lib/jpegli/README.md

See also https://giannirosato.com/blog/post/jpegli/

From my personal testing a straight conversion with no arguments has roughly halved the size of my picture collection (although admittedly they were likely not very compressed to begin with).

Is it possible to add this encoder under _exec along with jpeg2000 and jbig2?

[jbarlow83]

Could be done with a plugin, but I can't see taking this on myself any time soon.

The basic implementation is straightforward, but some of the ways that program reformats JPEGs would need some adjustment to work with PDF and testing time to confirm it works with all PDF viewers.

Perhaps the authors of cjpegli could be inspired to contribute their changes to the Python Pillow project. Then all Python software that works with JPEG would benefit.

[Earnestly]

Yes, the pillow point would be pretty wonderful. There doesn't yet seem to be much discussion within pillow about adding it.

Also just to be clear about this, cjpegli produces ordinary jpegs. It's not incompatible with readers or viewers that can already decode jpeg. Like mozjpeg, it's a drop-in replacement encoder.