This is where you went wrong. You say it’s nonfunctional - so why did you use it?
Per that link:
“Almost all pixel formats supported by JPEG XR can be loaded. Incompatible formats, however, will first be converted to a representation that GIMP understands (this means you’ll loose HDR data, for example). All RGB pixel formats are converted to 24bpp RGB, all RGBA formats to 32bpp BGRA, and all grayscale formats to 8bpp Gray. Black-white images are imported as indexed images.”
Note that it didn’t say you lost HDR metadata - it says (as is expected when trying to cram float16 into an int8) that you will lose data
If you really want to create a Rec. 2100 encoded 10bpp image from a JXR, you can start with the Python tool I linked, but it’ll need quite a bit of work since doing that was unnecessary for the original purposes of that script - it was designed to generate TIFFs that could be ingested into RawTherapee, so it doesn’t alter the data at all and tags it as linear with sRGB primaries, which matches every JXR file I’ve ever seen. You will need to:
- Calculate the appropriate matrix for converting an image encoded with sRGB primaries to Rec.2020 primaries, I’d suggest https://www.colour-science.org/ for that
- Apply that matrix to the float32 numpy array representing your image data after the script loaded an image
- At this point you should not have any negative numbers - if you do, you probably made a math error because it’s extremely unlikely (but possible) that the image contains colors outside of the Rec. 2020 colorspace. (negative float16/float32 numbers are a signal of either a bug or out-of-gamut colors)
- Take your image data that is now floating point linear with rec. 2020 primaries, and apply the HLG or PQ transfer functions
- Export as HEIC/HEIF with appropriate metadata - I can’t help you here. Or export as int16 linear Rec2020 TIFF then feed that to ffmpeg to create a video clip, see Experimenting with HLG and HDR10 for still images - #22 by Entropy512 for an example