Why JPEG XL?

Introduction

Images are a very common type of data, however the formats in which they are usually stored has not changed much since the 1990s. Today images are most commonly stored as either JPEGs or PNGs, and while there are some other formats, they all lack the universal adoption of these two formats. JPEG in particular has proven difficult to dislodge, as it is actually really good. However thirty years on, in 2022, we can do a lot better than original JPEG. There are two main formats currently vying for this position, JPEG XL and AVIF.

JPEG XL is a new format developed by the Joint Photographic Expert Group, a working group of the ISO, to replace 1992 JPEG. While this is the same organisation, the people behind it are different. JPEG XL uses a lot of the same ideas as the original JPEG (I've got some slides on it here), but with many new ideas and improvements as well. It also supports almost any feature you could desire in a raster image format, including high bit depth (up to 32 bits per channel), HDR, colour profiles, huge images, many channels, lossless, transparency, and even animation.

AVIF is another format trying to replace JPEG. It is based upon the AV1 video format that has been developed over the past several years, and is fundamentally just a single frame of AV1 video in a fancy container. It also supports many modern features, and has recently been gaining browser support.

Why do I think JPEG XL is better?

In terms of density, that is how many bits it takes to represent an image at a set quality, JPEG XL and AVIF are about the same. Depending on the type of image that is being compressed changes which one is better. JPEG XL tends to be better on photos, or at high qualities, while AVIF does better on drawings or at very low quality, but there isn't a huge amount in it. They are both a significant improvement over JPEG, but I'm going to come out on a limb here and say that most people don't actually care about getting images super small. A decently compressed JPEG is small enough for loading over mobile data connections, and network speeds are only going to get faster.

What I think makes JPEG XL better is its feature set. A JPEG XL file can store pretty much any conceivable image. Unlike the original JPEG, it supports high bit depth content, with accurate colours. And compared to formats like AVIF, which are based off video codecs, there are no arbitrary limits for which you are likely to encounter. AVIF has a limit of an image being less than ~16,000 pixels in any dimension, which large images can exceed, and most certainly will in future. While this can be worked around with tiling, it is much nicer if the format can just avoid having these issues.

Another thing against AVIF is its implementations not supporting the full set of its features. For example, AVIF support was recently added to Android in Android 12, but it doesn't properly support 12 bit per channel images. Firefox supports AVIF, but not its animated variant, which leads to some major websites being broken. A lot of these incomplete implementations come from hardware decoders being used for AVIF, as it is quite slow. JPEG XL is instead fast enough to use the CPU, which makes it more portable, and ensures that all the features are supported.

JPEG XL has a very interesting trick that I think will greatly aid adoption. It can losslessly transcode original JPEG files into smaller JPEG XL files, and then the original JPEG file can be retrieved if ever needed. This is possible because JPEG XL is a superset of everything that JPEG can do. This means that JPEG XL can improve density of the billions of existing images in the world, while with other formats you wouldn't want to touch them due to compounding loss.

Probably the biggest detractor from JPEG XL currently is support. AVIF has recently gained browser support in Firefox, which along with Chrome make it supported most places. JPEG XL is behind a flag in both of these browsers, and thus not enabled by default. I put this down primarily to JPEG XL being newer than AVIF, with the JPEG XL spec being published in 2021 while AVIF was back in 2019. I'm hoping that this will improve soon however, as version 1.0 of the libjxl reference implementation of JPEG XL should be coming within the next few months, which should hopefully encourage anyone waiting for API stabilisation, or who is scared off by the current v0.X status. You can use the below images to see if your browser supports either of these two newer formats.

Although I have been somewhat critical of AVIF in this article, I don't think it is a bad format. If you already have support for AV1 video in your environment, such as a browser, it is very easy to implement. It is also very good at making images look nice (though not necessarily accurate) at very low bitrates, when artifacts would make other formats untenable. I think AVIF could be a great format for CDNs to automatically transcode files for web delivery. I just don't see it becoming a long terms format, and instead I expect it to be replaced when the next big step in video formats comes around, much like webp now.

Overall I really want JPEG XL to succeed, as it would bring a whole load of modern image features in a single format that could then be the goto format for everything. In my view JPEG XL has the potential to do what JPEG did, and stick around as the dominant image format for decades to come, something I don't thing its competitors can do. But for now all we can do is wait and see what happens.