JPEG XL

this is silly

It's Adobe

yep, that sums it up

> re: using windows photos app nice

so my findings so far: `-jxl` by itself does nothing `-lossy` defaults to JXL unless you set `-dng1.4` or something to get lossless you do `-jxl -losslessJXL` `-jxl_effort` does nothing

First they add JXL to the DNG spec without thinking for a second about how to make the most of it, and then they make it nearly impossible to use properly

affinity photo just says compression type `JPEG` so im not sure what the heck it's doing

the compression ratio `40mb:98mb` seems indicative of JXL's lossless abilities

The ordering of the file it weird, so it's probably seeing the jpeg preview

https://discord.com/channels/794206087879852103/822105409312653333/1253000937669132410

but it could also be doing something stupid

There's some more info there about the horrors Adobe are doing to our baby

is there any way to """extract""" the raw data from a DNG into something like PPM so I can do some testing with `cjxl` instead of adobe's garbage?

> "Both types of JPEG XL bitstreams are supported in DNG" Why would you want to use JXL's containers in a DNG? Isn't that the point of the DNG header? Wouldn't that be redundant duplication? Is there something I'm missing here?

So many questions

You *can*, the problem is DNG is actually TIFF, which is tiled, which means cjxl would only be running on a handful of pixels at a time. Could extract a lossless JXL tile instead, since those were 672x752, then crop it down to 256x256 (Default group size) to simulate Adobe having braincells

You want to run exiftool on it, find the tile length and tile offset (1 value listed for each in a normal DNG, a list of values for a JXL encoded DNG), head to that decimal spot in a hex editor, then select the tile length as a decimal value too (there'll be matching pairs for JXL), save the selected bytes to a file and then you either have lossless jpeg or lossless jxl, I used ffmpeg to read the jpeg

So DNG is a fancy ZIP? lol

At least then it'd have some extra compression

Reject DNG, accept TAR.XZ 🤣 😜

can't be bothered doing it manually, im gonna write a C script or something

Should be simple enough, just find the tags, read the offset and then the length, save those bytes and then move onto the next pair until there's no numbers left

i wrote this C program to do it warning: the code really sucks also it only works on little endian systems but does support both little and big endian DNG files