#6367 adds support to import JXL images. It was opened a few years ago, but progress stalled because of missing dependencies. A PR to add export capability was never created because the same issues holding back import would have blocked export.
The dependency issue was recently resolved, so now anyone who’s interested can try the workflow artifacts.
Scroll to the bottom of the PR page. Click “show all checks”. Then click “Details” for the workflow that matches your operating system. On the left sidebar, click “Summary”. The download links should be at the bottom of the next page.
The PR workflows are put on hold whenever I push a change, so the downloads aren’t available until someone approves them. Another source of artifacts are the workflows from my working branch.
Feedback from anyone who’s using jxl import would be appreciated…
For anyone waiting for jxl export… that will be in a different PR. The import PR needs to be merged first because it updates the build scripts and workflows to include libjxl.
I can do that. Anything you’d like me to test in particular? I’m a little busy over the next few days, but will have plenty of time a bit later on. In general, eveything seems to work as expected; however, I tend to use RT to almost exclusively to work on RAW images.
@martbetz Whenever you have time would be great. If you have a github account, please comment at the PR. Anything you can think of to test would be good. Testing on multiple platforms would be nice, if you have access to more than one.
What should work is expanded bit depth (float) even when the original was 8bpc. My understanding is jxl uses float internally, regardless of target. This is potentially useful as a means of bit-depth expansion when the only original is a jpg (via lossless conversion to jxl). (Using jpegli would be more direct, but that would be for another PR.)
Metadata reading should be automatic via exiv2. It seemed to work on my computer (Linux), but would be good to know if there are problems on other computers or platforms.
Thumbnails should load. I don’t recall the method it’s using to create them. I’d prefer to defer attempts to improve performance for another PR, but could be helpful to know whether you notice any change in thumbnail loading.
ICC profiles are unlikely to work at all. This is a source of endless problems that should be deferred. But it would be interesting if code elsewhere in the pipeline somehow manages to apply profiles correctly.
