JPEG XL vs. AVIF for web use-cases
jakearchibald.com
jakearchibald.com
libjxl's slow speed at efforts >7 is primarily due to an iterative evaluation-optimization loop of a perceptual error metric called Butteraugli (by tuning quantization weights). This, as well as some other computations in libjxl encoding, happen to be data-parallel and run faster on GPUs than multicore CPUs.
In case anyone is interested, I've been working on a GPU-accelerated libjxl port (supporting CUDA and Metal for now) that makes this much faster, closer to libjxl's normal speeds: https://github.com/yunho-c/gjxl - geared for encoding a large batch images into JPEG XL. It's also faster by around 2.9x at normal effort (7).
I also remember the big claims about WebP's superiority, which ended-up in practice, at equivalent file-size, mostly trading away some of the JPEG artifacts that we all know for some unusual new ones.
I don't think there's a large-enough margin in this space that a new format could chew into to make the general public care: image compression seems like a solved problem overall. The second best thing is to come-up with a format that can make the content producers not have to care either, i.e. be the best in class or close-enough in a variety of use-cases (lossy, lossless, fast, animated, hdr, progressive, …). My naive impression was that jxl is mostly interesting in that regard.
RAW can be it's own entire nasty^nasty rabbit hole. The sorts of stuff that honestly do make me wonder if certain raw formats have their own unique ways of exposing vulnerabilities in accepted ways of implementing.
Things that come to mind are;
- Compressed RAW images. Both the lossless and lossy variants, esp because every manufacturer does their own thing, well except maybe MFT[0], buuuut;
-For MFT, my understanding is that the RAW doesn't just contain some info for LR/C1/DPL[1] or any other sufficiently implemented processing tool to apply correction data (again, I could be wrong, but I think that is something that the Micro 4/3 format does.)
[0] - Micro 4/3
[1] Light room, Capture One, DXO PhotoLab (FKA DXO OpticsPro)
MFT is a sensor size and lens mount, not a raw file format. Panasonic and Olympus (among others) have made cameras using this standard, but use different raw file formats.
JPEG engines are getting better and more configurable (Fuji recipes, LUTs on Panasonic) which makes this workflow more viable.
what i’d love is a good intermediary archival format that is more space efficient than RAW while keeping more editing latitude than a JPEG. using a dynamic preserving tone curve like Log in 14 bits and compressing it losslessly. although i wonder if that would be much better than compressed DNGs.
What improvement do you think is necessary for people to care?
Genuinely curious. Spoiler link with my answer to avoid any bias: https://lgms.nl/p/cau/?b64&ZVRNb9QwEL0j9T%2BMKqFtpe1Cqcq9qoT...
It means the technical differences between format, and the real-world choice of which file you want, don't match.
I have tried doing that with curl/wget but have been unsuccessful, websites still serve webp even when the original URL ends with .jpg or .png... Could you explain how you do it?
If you're sure cURL is sending the Accept header correctly without WebP, e.g. just "Accept: image/png,image/jpeg" (note my error in my earlier post, it's Accept not Accepts), and the server is still serving WebP, then it's misconfigured and there may be nothing you can do, short of swearing at it.
- The Firefox screenshot w/ lossless JXL at 66.9kB, when with a better encoder you get get it down to like 40.7kB (39% smaller) [1] (or 43kB if you need it to work with jxl-rs [2])
- The cat illustration w/ lossless JXL at 255.3kB, when with a better encoder you can get it down to 178kB (30% smaller) [3]
[1] https://www.linolevan.com/05-screenshot.jxl
Curious what is the better encoder here referring to?
What encoder?
Have you tried encoding with JPEG XL’s modular mode? By default, libjxl only uses it for lossless and I haven’t tried with jixel, but in my experience, it often handles clean edges and smooth surfaces (the likes of which you might see in digital art) better.
> I'm a little nervous about this, because discourse around JPEG XL can get a little… weird.
As someone who’s been rooting for JPEG XL for years now, I also find the discourse a little upsetting. When first introduced, the format was very competitive with AVIF and steadily improving, until Google decided to single-handedly halt web adoption (de facto, because market share). No careful evaluation presented, no solid argumentation given.
My understanding is that the future of the format became uncertain at that point, because web adoption is regarded as crucial and no one wants to go all-in on a codec that may not end up filling the universal role it was designed for. Of course, AV1 encoders kept seeing lots of interest and improvement in the meantime, because everyone was already all-in on AV1 for video.
And now, when JPEG XL has finally painstakingly clawed its way back onto its original trajectory, now there are comparisons popping up showing how AVIF actually compresses better. People end up quoting them around social media, painting the picture that JPEG XL turned out to be the inferior format after all. It hasn’t, it just got a massive handicap.
Not that it’s a problem with your article. It’s informative, you compare the current state of available encoders for a specific use case, and you mention how they’re still evolving. It just sucks that we can’t really do a fair comparison of the formats at this point, yet that’s exactly what people will end up doing anyway, and it’ll really suck if it ends up hurting JPEG XL even more than Google already has.
You perfectly summarise how I felt as well. The revisionist also tries to paint the story as JPEG XL was first banned due to security issues which is again not true.
But that is OK. Since JPEG XL has finally and rightfully won its place, mostly thanks for Safari. We can now hopefully wait for better encoder.
Or even better, participate in improving it. The reference encoder is, I believe, in the libjxl repository[1].
Much of the recent effort seems to be focused on the performance of the Rust version of the decoding library[2], which is quite understandable because of the fresh browser adoption.
[1] https://github.com/libjxl/libjxl
[2] https://jxl-rs-perf.lucaversari.it/?config=desktop%7Csingle&...
ffmpeg -i {input_file} -filter_complex "[0:v]format=pix_fmts=yuva444p,split[color][alpha];[alpha]alphaextract[alpha_chan]" -map "[color]" -map "[alpha_chan]" -c:v libaom-av1 -crf 25 -cpu-used 3 -still-picture 1 -aom-params "sharpness=2:enable-chroma-deltaq=1:arnr-strength=0" -color_range pc -colorspace bt709 -color_primaries bt709 -color_trc bt709 {output_file}https://github.com/aomediacodec/libavif
avifenc supports args like enable-chroma-deltaq. I've used those but recently stopped, and nothing substantial changed.
Otherwise the massive size reduction for storage and transfer speeds makes it incredibly attractive for modern web applications.
Granted, AVIF performs impressively for, say, a huge hero image for a blog where you're out to create an overall impression and people aren't going to look at it closely. It does not impress me when I am compressing photos I took with my mirrorless and I expect them to look like photos taken with a mirrorless and not cell phone snaps.
That's what I noticed anyway, not a car or photo guy either.
https://jakearchibald.com/2020/avif-has-landed/demos/compare...
I was talking about. He linked another one! Guess people finally realized that one sucks and now they have a new one. I screwed up.
You're looking at it in a browser that doesn't support JPEG XL, so it's falling back to a WASM renderer, and I think it handles resizing differently.
This is totally my bad. I thought I was doing a good thing by adding that fallback renderer, but it isn't identical and it's giving people the wrong impression.
The lossless version is JPEG XL, hence the difference to AVIF.
Superficially, that benchmark and this one seem to be taking similar measurements. They both measure encoded size versus encoding time versus perceptual quality metrics, although they present the data on plots with different axes. I haven't tried to reproduce the results of either, but I have no reason to doubt either benchmark.
I wonder what accounts for the difference. Maybe the libavif encoder got a lot better in the last 2.5 years. The older benchmark uses libjxl v0.10.0 and libavif v1.0.3, while the newer one uses libjxl v0.12.0 and libavif v1.4.2. Or maybe it's the set of images used. The older benchmarks aggregates results for small test sets of images, while the newer benchmark tests individual images. The newer benchmark does find an image where libjxl wins on file size (the weasel), but it still loses on encoding time, whereas in the older benchmark libjxl almost always won on both. Or maybe it's the hardware (M3 Pro vs. M4 Pro).
[1]: https://cloudinary.com/blog/jpeg-xl-and-the-pareto-front
In particular, I don’t think there was significant progress on compression between libjxl v0.10.0 and v0.12.0, and there some regressions from earlier versions still remain. The article also measures the quite recent jixel, which is derived from the older libjxl-tiny and apparently does better in many cases. So JPEG XL encoders definitely have lots of room for improvement. With JPEG XL finally gaining wider adoption on the web, perhaps they’ll catch up or get ahead again. AV1 encoders are also still being improved, though.
I am still on webp, thinking that I am on the bleeding edge in a world where artworkers are obsessed with pixels with JPG/PNG as the only formats they know, which is an advance on GIF.
Every browser advertises what formats it will accept and if it is in 'data-saving, low bandwidth mode', so I just configure the CDN to keep the names the same '.jpg' but to send the bestest format for the situation/browser, whether that be webp or AVIF, fetching super-high resolution files from the origin server, outputting them in 'retina' resolution with none of those 'picture' elements referencing half a dozen 'responsive' sizes.
In this way nobody has to care what image format is being served, and if it is deemed that everything is way too compressed, the CDN can be flushed of the cached images, quality increased, and cache rebuilt.
It looks like I will need to update my server configuration to use JPEG XL in the usual five years' time, given the lack of any revolutionary benefit.
I was at Google for 10 years, then went to Shopify for 2 years, but I missed the web. For most of the time I was at Google, it mostly did what was good for the web, rather than pushing Google's interests, but that started to change, which is one of things that led to me leaving. E.g. see what happened with the Prompt API.
I am always on Team Hamilton even though I appreciate that he is a little long in the tooth and doing the classic retirement act in a 'stupid red car'.
I am obsessed with SVG and image compression/delivery, and I would like it if you could introduce the concept of an 'SVG sampler' to the world. This is very much inspired by embroidery samplers from the times when women needed to sew.
The point of the sampler is to impress, so as to get the work in, however, it is also a reference, with stitches to copy for lettering, numbers, religious iconography, sock-darning, button-holes and whatever else needs to be repeated.
With an SVG sampler the idea is to use the whole SVG toolkit, filters, masks, clip paths, symbols, views, SMIL, CSS and everything else.
Sure this can be done in Inkscape, however, the goal of the SVG sampler is to make it human drawable. For example, the search icon is a circle and a line, which can be done in a 45Kb export from Illustrator, with the circle defined as a thousand points, specified to NASA numbers.
Or it can be done in ten lines or so of SVG primitives, with the circle and the line using simple two-digit integers. The former is not something that a mere mortal could draw, whereas the latter could be.
Clearly some paths do need to be complex and imported from elsewhere.
SVGOMG will be able to mash the 45Kb Illustrator export into a tenth the size, however, it will not be as concise or as human readable as the 'haiku of the SVG sampler symbol'.
The SVG sampler does not have to be web-optimised, it can max out the CPU for the PC to go full China Syndrome and melt its way through your desk. However, it could be forked, culled of CPU-eater show-stoppers, to just have the icons needed for the next project.
Another way of seeing the SVG sampler idea is 'SVG golf', where you want to strive for maximal 'use' clones, maximal human readability, yet also an insistence on detail such as light/dark mode currentColor things, animating pretty much everything and using the primitives rather than paths, even if the code is longererer according to SVGOMG.
I think that SVG falls within an awkward gap where artworkers don't care for the code with programmers also not caring for the code, because SVG is artworker stuff, not for programmers, right?
In a typical team workflow there is no scope for this to change, the frontend developer is obligated to use the 'assets' churned out by the art worker and the backend developer certainly doesn't care.
Hence, an SVG 'sampler' competition of sorts, with some acknowledgement, could help Firefox win over some developer mindset to FF. Already MDN has the de-facto best documentation for SVG, I probably have a dozen MDN tabs open right now, maybe with all of them on some type of FEconvolve type of feature.
For the sampler, the IDE rather than the graphics program is how it works. Anything with 'sodipodi' type of extra markup can be considered not worthy of being an SVG sampler. Really it has to be a mostly text-only thing, under version control, and not needing CodePen or other. To go with the SVG there needs to be a 'sampler.html' file to show the sampler symbols in 'inline mode' as well as CSS, the favicon and as shown as an 'image' element. Scripting is allowed, but really the goal is to keep everything in the SVG file, to not need huge amounts of extra files.
Ideally comments outside of 'title' and 'description' should not be needed, the author should be able to explain their own code, hence an emphasis on 'defs', sensible id names and descriptive ids.
Since F1 is your thing, you are welcome to give the SVG sampler idea a F1 inspired ethos to it. If you think of the FW14C with the active suspension and whatnot, or the F-duct, DAS and other wonders, then applied that design rigour to SVG, think how good it could be. Added marks for everything good looking, however, there needs to be this solid human-readable underpinnings.
Please do something for SVG at Firefox, to inspire people to get into it rather than assume it is someone else's job. Maybe just have a competition, no prizes needed, with folks able to vote on awesomeness, sophistication, haiku-esque simplicity and how educational a given sampler is.
I see no reason why you can't call yourself 'uncle Bernie', to cosplay as the benevolent dictator that gives out a star prize for an SVG sampler (every time there is an F1 race, rather than monthly, as that would be tedious).
As for AI, Google suggest does offer up SVG code that never works, it gets a duff namespace declaration on line one, dooming any of its concoctions to never open. But where we are going with 'SVG samplers', we don't need no AI robots.
After reading this article, there's something to point to prove I'm not going crazy.
This is something I've felt doesn't get enough attention when comparing JXL and AVIF. AVIF can be the replacement for GIF.
><video muted autoplay> is right there.
>No, I'm not adding a demo. I don't want it on my server. Use video formats for video.
I'm not sure if "Use video formats for video" is saying to use AVIF over JXL or saying to use AV1 over AVIF. If the latter, GIF-like animations have a different enough user expectation that it makes sense to have it as something different from regular videos. Even <video muted autoplay> doesn't encode the full expected behavior because GIFs are not expected to have controls like pause, play and scrub. And even if you hide the controls users will expect similar behavior when opening it in an image viewer or in a new tab. Images do not have controls so they fit better as an image, although maybe they need their own category.
Also, I didn't realize how good AVIF was at progressive rendering. I don't think I knew it could do that at all. I think AVIF is the clear winner for the web, although it is still nice to have JXL for lossless. Somehow I feel that AVIF still hasn't gained as much usage as it should, especially now that it is supported almost everywhere. I think I see more webp than AVIF, and JPEG is still everywhere. Maybe now that JXL is out, developers will take another look? AVIF is supported by 96.35%-98.83% of browsers according to caniuse whereas webp is 97.18%-99.68% and JPEG is 97.28%-99.78%.
I also didn't know there was a Rust JXL encoder named jixel.[0]
Though given the lagging speed in decoding, is there a faster tying JXL decoder planned?
Think of it as zip compression that is specially designed for JPEG and putting the JPEG data in a different/better container.
JPEG-XL seems to be the best solution for storing and distributing high-quality images. For myself, I want high-quality images, so JPEG-XL is what matters, both for new images and for compressing the old archives of JPEG images.
While I have my own Web server, it is for private use and not for commercial purposes, so I have no interest in using AVIF to achieve the smallest image sizes, but at a lower quality than I want.
And the article supports what I have said.
Quotation from the article:
> But JPEG XL catches up at very high quality. Around SSIMULACRA2 85, libjxl's default matches AVIF's default, and at maximum efforts, jixel 'extraslow' is ~6% smaller than AVIF s0. jixel 'slow' is 10–13% smaller than AVIF's default from SSIMULACRA2 80 upwards.
ssimulacra2 85 is rather close to median quality in the internet
ssimulacra2 90-95 is rather close to high quality in the internet
ssimulacra2 98+ is close to camera quality
JPEG XL is the ideal high end photography format. AVIF seems better for a space efficient web format.
For a wide variety of usecases, not for every isolated usecase. Today, AVIF is better for showing people ads^W^W^W web content delivery. This article is titled “JPEG XL vs. AVIF for web use-cases”. The other article recently discussed¹ is subtitled “Investigating JPEG XL's place as a Web image codec” and ends:
> JPEG XL isn't useless; it is genuinely compelling technology for use cases beyond the Web.
¹ https://news.ycombinator.com/item?id=49690554
(Either way, can we agree to take webp out behind the barn?)
JPEG XL has always been targeting high quality encoding metric first before the lower end low BPP images, which AV1 excels at with or without the last 6 years of improvements.
The current top comments about MP3 and LAME is again about how encoder extract maximum potential. JPEG XL was may be the best in 2021, but has been fighting for survival since then.
There a lot of different images that JPEG XL encode better. Screenshots with text is known to be not something they have optimised for or think it could be good at. Lossless mode may be better. Edit: As one of the other comment here proved my point.
Had all of these, even just a hint of AVIF improvement be told, may be the sentiments would have been a little different. But then again I am not one who would trust the same people who sold me WebP and AOM roadmaps.
But at least AV2 is coming. And JPEG XL is finally in every browser. Hopefully we can all have something even better in the future.
JPEG XL also has some big benefits for many use cases like lossless re-encoding and supporting higher resolution. AVIF images have a max size of 8192x4352 last I checked. That's not a problem for web images almost all of the time, but there's certainly uses where you might want more. I think AFIV might be limited to 12bit overall while JPEG XL can do 32bit. Again, is that something many will take advantage of on the web? No, but if it's something you care about it might be really meaningful to you.
Also, if you care about the 90% quality level, JPEG XL often outshines AVIF, but when I'm putting images on the web it's often well below that.
Some of it was bad information. I saw some people saying JPEG XL being way less computationally intensive (even less than WebP) and offering much better compression. Some of it was selective data - do you only test things at 95% quality level and declare a winner?
But I think the big thing is that a lot of people wanted the two formats to succeed or fail based on their merits. I think a lot of people don't like AVIF because it isn't well supported outside of browsers - and there can be a hope that the JPEG name would mean JPEG XL would get the kind of adoption JPEG did.
And it's hard to truly compare formats - or benchmark in general. As the article shows, different JPEG XL encoders offer vastly different performance. And when we're dealing with lossy compression, sometimes it can be a bit of personal preference. At some point, people started enjoying the way that MP3 garbled symbols (the instrument) and that was the sound they expected a symbol to make.
This isn't true btw. The format supports up to 64k x 64k, although browser decoders currently fail beyond 16k x 16k.
Not sure when you'd serve something like that on a web page though.
If I reduced the res a bit, then it worked, so there did seem to be a limit there somewhere. height and width are stored as u32s internally though, so yes, the format itself technically supports it...
I never looked into it more, so might have been a libavif issue, or maybe the way I was using it (maybe it has defaults or profiles?)
https://aomediacodec.github.io/av1-avif/#profiles
I'm assuming if both the encoder and decoder support more than the baseline profile, then higher resolutions up to 64k x 64k are possible, although I would worry the file would be incompatible with some software.
As the article points out, this might change yet again in the future, depending on how much AVIF keeps improving and whether JPEG XL development picks up steam again and how far it’ll go.
(also like the BPP size metric that is picture size-independent)
Wish that widget with improvements (that allows comparison of 3 pics with 3 buttons: original/codec1/codec2) were the widespread standard (also, "hold" is a rather unergonomic action)
Otherwise, seems like AVIF has had impressive gains in both speed and quality, wonder whether JXL could catch up
The images in the article max out at the maximum size I'd want them displayed.
Not to call into question the article’s author’s expertise, just to point out that they are a single individual and not an authoritative source (which you haven’t implied, but some people might read that way).
Try it: run cjxl › djxl and hash the output against the original.
Text has lots of edges, hard transitions between bright and dark areas. So text requires lots of high-frequency detail, and JPEG adds a lot ringing as it bounces off all those lines and curves.
AVIF seems to follow contours, so it does a good job with line art.
Otherwise, it's hard to beat the Mozilla JPEG encoder.
JPEG XL can represent lossless images that exceed the limits of human vision. I don't expect it to replace FITS for astronomical images, but the future is weird.
(Super huge thanks to Jake Archibald for the SVGOMG web app. Which is also fantastic for line art.
https://jakearchibald.github.io/svgomg/
Squoosh is a web app front end to various image encoders
)
Overall I am getting that AVIF is better on the web...unless Lossless is needed, in that case better take JPEGXL.
Adding formats to the browser gives us more code to maintain, increases the binary size, and increases the attack surface area. That's why, at Firefox, we insisted on a Rust decoder.
In that case, don't nanny everything as a monolith. Let formats/codecs can be installed as extension/plugin.
For example I really want to webrtc to f' off, I can't. I can't even disable it for a particular site like microphone permissions. It sucks!
And to my knowledge, webrtc stack is so much more complicated than an image format, so many sites trying to grab my private IP using webrtc. yet every browser vendor are happy to support it!
There are dozens of .wasm based image encoder/decoders outthere, no?
It's not like it could access network or IO, unlike extensions do, which could steal your data after authors sell their popular extension and buys profits from stealing data from installed users.
AVIF uses the AV1 codec. For the past several years, AV1 has been enjoying a wide interest and a continuous stream of improvements as the best, open, royalty-free video codec around that pretty much everyone had agreed to support from the get go.
JPEG XL started out with a smaller group behind it and took a massive hit when Google decided not to support it in Chrome (and by extension, the entire web, because market share). That made JPEG XL’s future as the universal format very uncertain for a few years. Excitement waned, development slowed down and was further divided to develop a new encoder which Mozilla later said they wanted.
So for a good while, AVIF was steadily improving and approaching its full potential, while JPEG XL was not. The article compares the current state of the current encoders; in no way does it indicate how good the formats themselves are. The article makes note of this. [2] It’s been known for years that JPEG XL compression can still be improved by quite a bit, so I imagine the current encoders are still far from reaching the format’s full potential. (Heck, they don’t even make use of all its features from what I know!) Back when it was first proposed, JPEG XL was very competitive. With interest in the format hopefully renewed, we might see it catch back up with AVIF and overtake it again. Or we might not. We have no way of knowing for sure right now.
[1]: https://news.ycombinator.com/item?id=50128587
[2]: https://jakearchibald.com/2026/jpeg-xl-vs-avif/#:~:text=Both%20AVIF,in%20futureIn full screen mode, the "hold for lossless" version "Firefox help"/"Submit feedback" are not underlined. In inline mode they are. Both modes seem to render a scaled down version of the image, which seems to cause these problems.
It does seem like AVIFs win on small images, and JXL gives you progressive rendering, easy upgradability on JPEGs, and better lossless and high bitrate images.
It does sound like the remaining areas jpegxl is better at are almost all non-web use cases.
For me personally, AVIF is still better, for various reasons (not solely file size compression ratio as such, though this is my number one go-to objective; browser support is also useful but to me is not so important as it is).
I am keeping a close eye on JPEG-XL, but right now I am quite settled on .avif. All my local collection is in .avif, when before it was a mixture of .jpg, .png and a few animated .gifs. From my findings a few years ago, avif is superior to jpg, png and gifs on pretty much everything it matters. On very high compression, though, I also noticed artefacts, but we had that even more so with jpg, and for 1:1 comparable file size, avif beats jpg with both eyes closed.
By the way, great article. I hope he can do another review, based on that blog entry, in, say 3 years. Would be interesting to see how avif and jpeg-xl have improved. In all likelihood I will still keep on using avif, but if jpeg-xl is a real winner in some areas, I may use it. (Lossless is one area where jpeg-xl excels, but I don't need to store satellite data; I want to store old family pictures in an efficient manner, and here it seems avif is better.)
I put together this Hammerspoon spoon with a few brew libraries and some inline Swift code.
You paste twice quickly to trigger it, and you can iterate formats with a keystroke. https://gist.github.com/thadk/8b178aa93dafcab92ca579825f8191...
IE never supported AVIF. Edge even shipped AVIF later than Chrome. One of the leaders of JPEG XL worked at Google until relatively recently. A massive amount of JPEG XL development has been done by a team at Google.
So, is animated AVIF only keyframes / p-frames, where each frame is independent from the other? I was under that impression, uncertain if there are any other capabilities it might have (like only updating part of the canvas when only parts change, like GIF supports)
That said, I last benchmarked over a year ago, maybe I should re-evaluate. My experience was that AVIF was too slow for on-the-fly encoding, while JXL was absolutely fine and JPEGLI was also pretty useful.
JPEG XL can't be decoded in such a way, and tons of people still use JPG.
What optimizations are being added? I thought it was basically the exact same codec, but different packaging and usually some adjustment to how the encoder is tuned.
Amazing.
JPEG XL doesn't have that benefit, while sites adoption is arguably a worse issue than browser adoption.
To demonstrate the point even for AVIF itself - it has been in the browsers for a while. Can you use it on common sites? Good luck if you can, even major sites like GitHub and GitLab have issues with it due to some site code depending on hardcoded list of known formats or other similar problems.
ISO-BMFF container is bloated and needlessly complicated. AVIF decoders didn't even want to reuse existing MP4 libraries for it due to all the baggage it brings, and implemented their own minimal parsers.
AVIF would have been better off with a simple wrapper similar to RIFF/PNG or extending AV1's own OBU header (which for inexplicable reason gets its data duplicated inside the HEIF container and then implementations are required to verify that both copies agree!)
I hope AVIF will skip AV2. We don't need yet another almost-identical format to migrate to.
With no codec for a new compression algorithm, it does not matter if the browser can read the container, you still see the "broken image" icon on your display.
You as user can then upload AVIF images to wherever images on the site are processed. And after that, let's say when newer codec inside AVIF will get a good browser coverage, you'll be able to start using it without waiting another decade for GitHub to pipe it through their site code.
For formats without a container, it's always a two step process. Browser coverage first, then another huge period of sites support catching up.
> You as user can then upload AVIF images to wherever images on the site are processed
Those image upload pipelines almost always process the image in some way (downscaling, compressing). So a new codec in the same file format can only break them.
But even if it didn’t, I can’t help but view this as a dependency hell. Somebody uploads a version of AVIF not yet supported everywhere. Things break. User doesn’t understand why, because user doesn’t understand that .avif != .avif. Developers have a hard time figuring out why, because they have to check all software touching those files and check what specific codecs or version those support for AVIF. And all they can do anyway is to wait for all affected libraries or software to roll out the new codec.
So I guess in other words, the codec is the file format.
Many other image formats are. AVIF uses HEIC, the same container that HEIF does.
> when AV2, AV3 etc. will come out, there won't be a need to go through another very difficult phase of sites adoption - it will still be AVIF just with newer codec inside, and it will be just up to browsers to catch up on newer versions and newer codecs will address technical needs as times goes.
Fingers crossed, but AV2-based AVIF specs haven't even been finalized, so who knows.
2. Is it possible to encode AVIF with hardware acceleration?
When it comes to hardware encoders, the nicest thing you can say about them is they're fast. Software tends to do way better at quality. Hardware AV1 encoders will be tuned for video rather than stills.
It's hard to rigorously judge a codec from a handful images, especially given the large impact they would have (with a widely adopted codec encoding bazillions of images in the coming years).
* There are many intricate details that go into codecs and visual perception. For instance, our sensitivity to visual errors are dependent on colors (all of hue, saturation, and brightness), spatial frequency of artifacts, local contrast and texture presence, et cetera.
And like OP mentioned, perceptual error metrics (including SSIMULACRA2) aren't perfectly representative of human-evaluated visual quality. For instance, libjxl deliberately makes choices that hurt SSIMU2 scores, but improve human-rated quality.
But yes, you can see this with jixel, which I'm really impressed by, but it's so tuned to SSIMU2 that it looks really bad to human eyes in some cases compared to libjxl.
Maybe the first step to designing better encoders is to build an extremely accurate visuoperceptual error metric. Seems like ML's rapid progress haven't caught onto it just yet.
Seems like others were curious too, given the Wikipedia page on Guetzli mentions:
> Butteraugli. ... How the hundreds of parameters that model the properties of the human visual system were derived remains unexplained.
My rough understanding is its multi-scale convolution architecture was inspired by biological visual perception (perhaps how the layers of visual cortex gradually represent higher level features, ish), and involved some sort of empirical parameter tuning based on human evaluations. It'd also be a very nice resource for beginners wanting to learn more about image/video codecs (like me :)
But that is correct and why any proper study should first train the participants on the type of artifacts and trade-offs. Like the reference to JXL choices: think its creator mentioned that AVIF might look nicer for faces because it was "smoothing" them but it was a worse representation of the original. So the study participants would need to be conscious of that trade off and be clear in terms of what they're asked to evaluate - whether the result looks nicer or whether it represents the original better.
Perhaps we can use clever experiment design to address it, like doing a "full-reference" evaluation (e.g., pick most similar two images, out of encoder A, encoder B, and original) rather than strictly A/B.