Comment by wizee
3 months ago
JPEG-XL provides the best migration path for image conversion from JPEG, with lossless recompression. It also supports arbitrary HDR bit depths (up to 32 bits per channel) unlike AVIF, and generally its HDR support is much better than AVIF. Other operating systems and applications were making strides towards adopting this format, but Google was up till now stubbornly holding the web back in their refusal to support JPEG-XL in favour of AVIF which they were pushing. I’m glad to hear they’re finally reconsidering. Let’s hope this leads to resources being dedicated to help build and maintain a performant and memory safe decoder (in Rust?).
It's not just Google, Mozilla has no desire to introduce a barely supported massive C++ decoder for marginal gains either:
https://github.com/mozilla/standards-positions/pull/1064
avif is just better for typical web image quality, it produces better looking images and its artifacts aren't as annoying (smoothing instead of blocking and ringing around sharp edges).
You also get it for basically free because it's just an av1 key frame. Every browser needs an av1 decoder already unless it's willing to forego users who would like to be able to watch Netflix and YouTube.
I don't understand what you're trying to say. Mozilla said over a year ago that they would support JXL as soon as there's a fast memory safe decoder that will be supported.
Google on the other hand never expressed any desire to support JXL at all, regardless of the implementation. Only just now after the PDF Association announced that PDF would be using JXL, did they decide to support JXL on the web.
> avif is just better for typical web image quality, it produces better looking images and its artifacts aren't as annoying (smoothing instead of blocking and ringing around sharp edges).
AVIF is certainly better for the level of quality that Google wants you to use, but in reality, images on the web are much higher quality than that.
And JXL is pretty good if you want smoothing, in fact libjxl's defaults have gotten so overly smooth recently that it's considered a problem which they're in the process of fixing.
> I don't understand what you're trying to say. Mozilla said over a year ago that they would support JXL as soon as there's a fast memory safe decoder that will be supported.
Did they actually say that? All the statements i've seen them have been much more guarded and vauge. More of a, maybe we will think about it if that happens.
2 replies →
I disagree about the image quality at typical sizes - I find JPEG-XL is generally similar or better than AVIF at any reasonable compression ratios for web images. See this for example: https://tonisagrista.com/blog/2023/jpegxl-vs-avif/
AVIF only comes out as superior at extreme compression ratios at much lower bit rates than are typically used for web images, and the images generally look like smothered messes at those extreme ratios.
Not everything in the world is passive end-of-the-line presentation. JPEG-XL is the only one that tries to be a general-purpose image format.
If that's the case, let it be a feature of image editing packages that can output formats that are for the web. It's a web standard we're talking about here, not a general-purpose image format, so asking browsers to carry that big code load seems unreasonable when existing formats do most of what we need and want for the web.
3 replies →
Even though AVIF decoding support is fairly widespread by now, it is still not ubiquitous like JPEG/PNG/GIF. So typically services will store or generate the same image in multiple formats including AVIF for bandwidth optimization and JPEG for universal client support. Browser headers help to determine compatibility, but it's still fairly complicated to implement, and users also end up having to deal with different platforms supporting different formats when they are served WebP or AVIF and want to reupload an image somewhere else that does not like those formats. As far as I can tell, JXL solves that issue for most websites since it is backwards-compatible and can be decoded into JPEG when a client does not support JXL. I would happily give up a few percent in compression efficiency to get back to a single all-purpose lossy image format.
Even Google photo does not support avif.
It's almost as if Google had an interest in increased storage and bandwidth. Of course they don't but as paying Driver used I'm overcharged for the same thing.
8 replies →
The killer feature of JXL is that most websites already have a whole bunch of images in JPEG format, and converting those to JXL shrinks them by about 30% without introducing any new artifacts.
> Mozilla has no desire to introduce a barely supported massive C++ decoder for marginal gains
On a slightly related note, I wanted to have a HDR background image in Windows 11. Should be a breeze in 2025 right?
Well, Windows 11 only supports JPEG XR[1] for HDR background images. And my commonly used tools did either not support JPEG XR (Gimp fex) or they did not work correctly (ImageMagick).
So I had a look at the JPEG XR reference implementation, which was hosted on Codeplex but has been mirrored on GitHub[2]. And boy, I sure hope that isn't the code that lives in Windows 11...
Ok most of the gunk is in the encoder/decoder wrapper code, but still, for something that's supposedly still in active use by Microsoft... Though not even hosting their own copy of the reference implementation is telling enough I suppose.
[1]: https://en.wikipedia.org/wiki/JPEG_XR
[2]: https://github.com/4creators/jxrlib
Another JPEG XR user is Zeiss. It saves both grayscale and color microscope images with JPEG XR compression in a container format. Zeiss also released a C++ library (libczi) using the reference JPEG XR implementation to read/write these images. Somehow Zeiss is moving away from JPEG XR - its newer version of microscope control software saves with zstd compression by default.
"Marginal Gains"
Generation Loss – JPEG, WebP, JPEG XL, AVIF : https://www.youtube.com/watch?v=w7UDJUCMTng
Marginal gains over AVIF.
(Also I am highly skeptical of the importance of these generation loss tests.)
2 replies →
>avif is just better for typical web image quality,
What does "typical web image quality" even mean? I see lots of benchmarks with very low BPPs, like 0.5 or even lower, and that's where video-based image codecs shine.
However, I just visited CNN.com and these are the BPPs of the first 10 images my browser loaded: 1.40, 2.29, 1.88, 18.03 (PNG "CNN headlines" logo), 1.19, 2.01, 2.21, 2.32, 1.14, 2.45.
I believe people are underestimating the BPP values that are actually used on the web. I'm not saying that low-BPP images don't exist, but clearly it isn't hard to find examples of higher-quality images in the wild.
Can AVIF display 10 bit HDR with larger color gamut that any modern phone nowadays is capable of capturing?
> Can AVIF display 10 bit HDR with larger color gamut that any modern phone nowadays is capable of capturing?
Sure, 12-bit too, with HDR transfer functions (PQ and HLG), wide-gamut primaries (BT.2020, P3, etc.), and high-dynamic-range metadata (ITU/CTA mastering metadata, content light level metadata).
JPEG XL matches or exceeds these capabilities on paper, but not in practice. The reality is that the world is going to support the JPEG XL capabilities that Apple supports, and probably not much more.
if you actually read your parent comment: "typical web image quality"
41 replies →
Wanted to note https://issues.chromium.org/issues/40141863 on making the lossless JPEG recompression a Content-Encoding, which provides a way that, say, a CDN could deploy it in a way that's fully transparent to end users (if the user clicks Save it would save a .jpg).
(And: this is great! I think JPEG XL has chance of being adopted with the recompression "bridge" and fast decoding options, and things like progressive decoding for its VarDCT mode are practical advantages too.)
> (in Rust?)
Looks like that's the idea: https://issues.chromium.org/issues/462919304
> and generally its HDR support is much better than AVIF
Not anymore. JPEG had the best HDR support with ISO 21496-1 weirdly enough, but AVIF also just recently got that capability with 1.2 ( https://aomedia.org/blog%20posts/Libavif-Improves-Support-fo... ).
The last discussion in libjxl about this was seemingly taking the stance it wasn't necessary since JXL has "native HDR" which completely fails to understand the problem space entirely.
The JXL spec already has gainmaps...
Also, just because there's a spec for using gainmaps with JPEG doesn't mean that it works well. With only 8 bits of precision, it really sucks for HDR, gainmap or no gainmap. You just get too much banding. JXL otoh is completely immune to banding, with or without gainmaps.
> With only 8 bits of precision, it really sucks for HDR, gainmap or no gainmap. You just get too much banding.
This is simply not true. In fact, you get less banding than you do with 10-bit bt2020 PQ.
> JXL otoh is completely immune to banding
Nonsense. It has a lossy mode (which is its primary mode so to speak), so of course it has banding. Only lossless codecs can plausibly be claimed to be "immune to banding".
> The JXL spec already has gainmaps...
Ah looks like they added that sometime last year but decided to call it "JHGM" and also made almost no mention of this in the issue tracker, and didn't bother updating the previous feature requests asking for this that are still open.
1 reply →
> performant and memory safe decoder (in Rust?).
Isn't this exactly the case that wuffs [1] is built for? I had the vague (and, looking into it now, probably incorrect) impression that Google was going to start building all their decoders with that.
[1] https://github.com/google/wuffs
WUFFS only works for very simple codecs. Basically useless for anything complex enough that memory bugs would be common.