scripts:media:batch_convert_photos_to_jxl

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

scripts:media:batch_convert_photos_to_jxl [2023-01-15 23:23] – created Tonyscripts:media:batch_convert_photos_to_jxl [2023-01-15 23:23] (current) – Tony
Line 1: Line 1:
 +====== Batch-convert files to JXL ======
 +
 +JPG and PNG are old. JPEG-XL is the hype new image format. It has many cool features which I won't exhaustively list here. [[https://jpeg.org/jpegxl/]]
 +
 +If you decide to convert your files, you will find cjxl can be slow on the slow presets. Also, compressing an image with multiple threads reduces compression ratio.
 +
 +So, it is typically better to compress each image with 1 thread, which gets better compression as no data needs to be shared, and just compress more images at the same time. Ie: instead of compressing 1 image with 8 threads, compress 8 images with 1 thread each.
 +
 +Note, this means you are going to use massive amounts of RAM with slow presets. You have been warned.
 +
 +First, get the list of images you want to compress. A simple way is 
 +
 +  find -name "*.JPG" | tee /tmp/files
 +  
 +cjxl can work with either jpg or png as input (use imagemagick if you need more).
 +
 +Next, I use GNU Parallel to batch process these:
 +
 +  parallel --ungroup -j16 -v -q -a /tmp/files cjxl -e 9 --num_threads=0 --lossless_jpeg=1 -d 0 "{}" "{.}.jxl"
 +
 +  * -j16 indicates this will run 16 files at once.
 +  * -e 9 indicates maximum compression effort (see man cjxl for more)
 +  * -d 0 indicates lossless mode. By default, cjxl is lossless for JPG inputs, and lossy for PNG. (-d 1 is imperceptible typically)
 +  * --lossless_jpeg=1 enables the super cool JPEG recompression mode, where JPEG files can be losslessly converted to the new encoding while saving 30% space. (no effect on PNG inputs)
 +  * "{}" will match the input file name
 +  * "{.}" is the file with the extension removed (which we add .jxl to)
 +
 +While CJXL preserves some metedata, it is not the complete bundle if you use something like ''digikam'' to manage photos.
 +
 +You can use exiftool (or probably exiv2) to re-write the metadata either from the image or XMP files.
 +
 +  parallel --ungroup -j24 -v -q -a /tmp/files exiftool -overwrite_original -TagsFromFile "{}.xmp" "{.}.jxl.xmp"
 +
 +**CAUTION!**  At the time of writing, exiftool can NOT write to JXL files - it will, but it can only be read by exiftool. Use XMP files or another mechanism for the time being!