JPEG XL

This channel is to introduce yourself – not for chatting. To chat, you can go to <#794206087879852106>, <#806898911091753051>, or whatever channel is relevant.

I'll start. I'm Jon, born in 1981 in Leuven, Belgium, still living in Belgium. I have 3 daughters aged 5, 7 and 9. For fun I play the piano and the drums. I got interested in image compression about 10 years ago, and made the FLIF lossless image codec (together with <@!799692065771749416>). Five years ago I started working for Cloudinary, doing nothing but image processing and image compression stuff. When I heard that JPEG was making a call for proposals for a next-gen image codec with the silly name "JPEG XL", I simplified and extended FLIF into FUIF, and submitted it as a proposal. Out of the 7 proposals that were submitted, FUIF and Pik were selected as a starting point for JPEG XL. The past 2 years we have been busy finding ways to combine the two codecs into a single codec, and I think we succeeded.

Hi, I'm <@!224363555074342912> , I first started getting interested in encoding when I started playing around with webp and later FLIF. I live in central North America, but otherwise I'm still learning CS and programming and being a (mostly) anonymous person on the internet. My favorite encoding tool probably is Squoosh (thanks <@!228116142185512960> ).

Hi, I'm Pieter, born in 1984 in Leuven, Belgium, currently living in Mountain View, CA, USA. Long ago at university I shared an office with <@794205442175402004>, and we worked for a while on a side project for image compression (which Jon would later turn into FLIF, though my contributions stopped here). After getting my degree, I started working for Google as a site reliability engineer in 2012. I got into contributing to Bitcoin in my free time, and in 2014 I helped co-found Blockstream with the goal of being able to work on it full time. Since last year I work for Chaincode Labs, a Bitcoin research and development center. I haven't actually ever tried JPEG-XL, but promise I will 🙂

Hello, my name is Stefan, born in 1987 and I now live in Bucharest, Romania. I work as a Linux System Administrator and I'm transitioning to automation. In my spare time I watch some movies and play some computer games. I was really passionate about Counter-Strike 1.6 and had a server for a while. I am interested about compression and was impressed by the features offered by FLIF and followed from the shadows since then

Ooh, counter-strike! That has been... I have no clue, 16 years since I played that?

Now it has been about 30 seconds since I did.

Hoi! I’m Sami, from 1993. In 2015, I migrated from France to Switzerland to join Google in Zürich, and in 2018 I joined the team that was working on PIK and now on JPEG XL. My hobbies include photography, kind of re-awoken by my work on image compression, as well as music, especially in the style that Tangerine Dream did in the mid-70s (nowadays, it’s bands like Redshift or Free System Projekt that are carrying the fire).

Hi all! I'm Andrew Shieh, from Sunnyvale CA. I run cloud ops for SmugMug/Flickr. We're responsible for many billions of priceless images and I'm working on (my secret project) getting our teams interested in jpeg xl. We've done a good amount of technology work around maximizing image performance and reliability while minimizing costs. Particular areas of interest are AWS (esp. S3), lossless compression, GPUs, color. Personally I'm most excited about the improvements in color spaces/HDR. I'm looking forward to the day when I can buy a DSLR/mirrorless camera that fully supports jxl so I can mostly stop using those unwieldy RAW files.

Hi, I'm Jonathan from Melbourne, Australia. I'm studying mechatronics engineering and I'm currently working at the Melbourne Space Lab on an imagery system for some cubesats that we are flying in the next few years. We're using JPEG XL as we have very limited bandwidth, we're particularly excited about using progressive encoding/decoding to allow us to continue downloading an image until we get to a quality that we're happy with. (Our downlink speed is something like 2kB per 90 minutes!)

SMD

Ahoj, I'm Petr, 43, son of Czech parents, citizen of the world. Fan of raster formats since the 1990s. Once read the full spec of PNG, interested in PNG optimization. Contributor to allrgb.com (look for "interlacing"). Author/maintainer of several articles about computer graphics (including JPEG XL) on Czech and English Wikipedias. Learned Python to be able to design new levels for a 1984 video game. Translator of Czech, Esperanto, English. Proofreader of Czech.

Hi, I'm Jordi. I've been casually interested in compression for a while now. I'm currently researching better compression schemes for databases in my free time. I've been following AVIF for a while, and I stumbled into JXL from there. I ended up following Jyrki on Twitter and from there, found this server.

Hi! I am just a pseudo-random graphics programmer/technical artist living in Germany (name's Aie and I am way too young), who is trying to make a tool for easy art palette generation using perceptual color spaces! I have been researching a lot of color spaces for that and discovered that this Discord has a lot of color scientists and that is really nice, because I can study a lot of important stuff here. Also, if XYB gets a polar form one day, that would be interesting for me to research as well. I feel like my brain was procedurally generated to crave perfect colors but have no color intuition, so color science and all the stuff associated with it seems like such a lifesaver for me. I hope I compute. Also, I got here because of Mr. Alakuijala spilling Discord links everywhere like the statistical chaos of experimental color data available.

hi there, from the sizecoding communityy here (making graphical intros in 256 bytes or less). interested to learn more about jpeg-xl

hello, just found this through the demoscene, looks neat

Hello! A new compressed image format complete with its own subreddit and Discord server intrigued me, which is why I'm here.

Hi, I’m Gilles, 23, from France (near Tours, where we have all those nice castles). I’m an apprentice in embedded systems engineering. I really like technology and regularly do different projects (https://gildev.dev/en/projets) around programming and electronics. Something I did a few years ago that I think is quite cool-looking is this breadboard computer: https://gildev.dev/en/projets/gdbc (unsure how fast it will decode JPEG XL) I learned about JPEG XL at the end of last year, read quite a bit about the subject and I really liked this format and its ideas. After participating a bit on the subreddit and following the GitLab activity, I just found out about this Discord (crazy I didn’t find it before, I got there through a 2 month old comment from <@!794205442175402004> on Bugzilla). Thank you everyone for working on this and *making the web (but not only) a better place*!

La france en force 😄 (jvais lurker tes projets :p)

hein quoi on est combien

Au moins trois si on compte les QC.

Je ne suis pas français mais je peux me débrouiller en français

(j'habitait à Bruxelles entre 2010 et 2018)

We need a french chan, because we are vip 😄

Old but nice t-shirt

I really miss your old hair 😄

it grows back. I cut my hair once a year, when big hair starts to get a bit too warm, and it grows back so by winter I have my natural head insulation again.

> it grows back. that's a cool feature

one day i searched 'what is the png equivalent is for videos' cuz png is lossless but compressed and u have flac for audio but theres no supa quality video formats right, so then logically i searched for animated png which led to jpeg video which was a real format in fmvs in playstation games i think and then that led me to jpeg xl which is a very promising project very cool and pog! i will use it on my website and for storing images when it becomes adopted

welcome! and in case it is still useful for you and you haven’t come across it, the video equivalent for videos could be ffv1, or else lossless h.264/h.265/avif

ffv1 somewhat influenced jxl btw. FLIF took some inspiration from ffv1, and some of the MA context tree properties in modular jxl originate from how ffv1 defined its (fixed) context model.

Right, I remember that!

Hello, I am Denis, I am interested in compression (among many other things). I am having a blast with your jxl-from-tree tool 🎨 . I am from France 🇫🇷

Bonjour et bienvenu!

Cool 😄 montre nous ton talent 😜

hello world! I've been following this project pretty closely from the technical side since there was just FLIF. I have been enjoying the lossless jpeg transcoding feature in my personal use. DSP is my bread and (cashew) butter. I've worked on the linear prediction of the SILK part of Opus a long while ago and also the chroma from luma prediction on Daala (which has been rolled into av1 and avif now). Also worked on some shader code for libretro/retroarch - huge fan of ICtCp I want to see jxl take over the world and replace tiff/(a)png/jpeg/gif/etc and be the one to rule them all. The three best things about jxl are its interoperability with legacy jpeg, its features centered around high-fidelity, and the flexibility of the format (layers, multiple channels). Truly forward-thinking design and some clever engineering - I'm all about that, haha Apart from that stuff, I'm a melophile, I like e-bikes, I do martial arts, and I'm into fashion. glad I found this server!

Welcome!

Hello! I'm Raphael and I'm currently not satisfied about the state of JPEG encoders/transcoders in Rust. Somehow I landet here and maybe I will release a nice Rust library for JPEG soon, if my ADHD lets me do that...

JPEG, or JPEG XL? 😉

JPEG 🙈 I'm new to picture/video encoding and JPEG looks like an easy start.

Just create a mozjpg C API for my Rust Library using cargo-c <:kekw:808717074305122316>

No

I'm Jyrki Alakuijala. I designed neurosurgical robotics and AR the first three years of my career, then 13 years of radiotherapy planning software (IMRT, visualization, volume modeling, radiation physics), now 15 years at Google, with last 10 years in data compression. I manage the JPEG XL effort at Google. I invented the XYB, butteraugli, gaborish, delta palettization, entropy clustering approach, brotli, the codification of 8x8 dc first in a system having variable sized integral transforms. I convinced the JPEG committee to accept lossless jpeg recoding as a new requirement for JPEG XL. I was the first to propose collaboration of FUIF and Pik as the basis of JPEG XL -- instead of competing like we did with WebP lossless/FLIF times. Before JPEG XL I designed WebP lossless (that we are now competing against), Brotli, the first version of Brunsli (Zoltan took it over after the first version), the main approach in Guetzli. I like to play the piano, sing in a choir, play silly computer games, learning to play the organ.

we could build a faster guetzli based on principles in JPEG XL

So you're several times more qualified than I'll ever be at anything <:PepeSad:815718285877444619>

Brotli? <:FeelsAmazingMan:808826295768449054> <:megapog:816773962884972565>

https://datatracker.ietf.org/doc/html/rfc7932

That's really impressive, think of how many megabytes that's saved online 👍

I'm not one of those super-smart people, just working really hard and surrounding myself with people smarter than I am.

https://tenor.com/view/doubt-press-x-la-noire-meme-x-button-gif-19259237

Hello, I'm a media file format nerd

Hi! I recently learned about JPEG XL and got interested pretty quickly. I am a Rust fan (since before Rust 1.0) and I couldn't find any pure Rust implementation being developed. So I started one https://github.com/vojtechkral/jxlrate (almost nothing is implemented, just the SizeHeader: https://github.com/vojtechkral/jxlrate/blob/main/src/decoder.rs, I really only spent a few hours on it so far). But today I learned about https://github.com/libjxl/jxl-rs/ by (presumably) <@!179701849576833024> which google had failed to tell me about before. So now I'm wondering if there's much point in continuing my effort. I'm not an expert in image compression. I should probably abandon that and help with `jxl-rs` instead. Let me know what you think 🙂 In any case, I'm glad this cool effort to improve JPEG is underway!

https://discord.com/channels/794206087879852103/804324493420920833/850460048327966770

Happy about any help ;)

<@!179701849576833024> Do you plan to keep the library 100% safe Rust? (That was my intention.)

We should probably switch this to PMs ;) anyway, generally yes but there might be a couple of performance sensitive places

I'll try to help with jxl-rs too, not an expert in image compression either, but I'll see what I can do 🙂

honestly you can do a lot with safe rust if you figure out how to make the autovectorizer happy one time i spent like 2 days painstakingly writing a simd version of an algorithm, then in like 10 minutes i wrote a safe rust version that just explicitly splits it into 128-bit large chunks before doing any work on it and the output was near identical

I've just found this golden piece 😃 https://cliffle.com/p/dangerust/

this is still off-topic for this thread 😛 but autovectorization is not *that* good, and it's not the only problem for performance anyway

Hello, I have recently come upon jpeg-xl and I was thinking about using it on some of my personal go projects and for that I'm going to start writting a go wrapper, where should I start looking on the API? is there an official API doc? (haven't found anything yet)

we used to have some doxygen-generated doc

not sure where it is now

you can look at things like https://github.com/libjxl/libjxl/blob/main/examples/decode_oneshot.cc and https://github.com/libjxl/libjxl/blob/main/lib/include/jxl/decode.h

i am tim

hi tim.

must click discord links Hey, I'm Ace and I don't really have a good reason for being here

we’ll help you find one

Lol

im back and <@!219525188818042881> is scared <:kekw:808717074305122316>

Of course I am, you're scary

🤔

i still have no idea why u say i'm scary

bc you're scary

Yes, others agree with me

I'm BarryCap

oh hi, I don't think I ever introduced myself I've been following development since the days of the FLIF slack, c. 2017 mathematician background I'm a pure algebraist/logician, or an applied analyst with not much in between

Hi, I'm just a guy who thought the "responsive" mode of FLIF was pretty neat, then discovered JPEG XL and have tried selling all my friends on it (And at least one agrees with me that it's pretty neat)

Also <#887993694593093653>

hi everyone. i'm a bit of an artist, programmer and occasional math enthusiast. jxl seems very cool and i though i'd drop by and maybe ask some stupid questions 😅.

Wow for some reason was not expecting all you people here... long time no chat everyone lol AV1 squad

I was JXL squad before I was AV1/AWCY squad

Hello everyone... I'm new here and got into the IQA and image compression topic through researching for and writing my thesis. 🙂

hello! just like to check out new codecs and talk about edge cases that i may find maybe

Hello! JXL is great! I'm particularly interested in very fast lossless compression (the current "lightning" mode is too slow for me), and compression of 12/14/16 bit Bayer or mono images.

I thought jxl seemed neat. I joined bc i was trying to compress an image by hand for fun

compress an image by hand in jxl? seems very tricky to do - I can barely decompress an image by hand 🙂

Could you encode/decode a JXL by hand on a livestream? It would be interesting.

Hi, learned about JXL a while ago but finally got a boost to try and conquer the art aspect of it, and I'm here to learn about that :)

Hyped about JPEG XL since one of the early open builds, used to do quite a bit of compression testing with it on my own, done some digging on the techniques used in it... maybe I could be considered an early adopter except I don't really use it that much in practice because of interoperability issues :/ also an enthusiast in multimedia compression techniques in general, love to study classic and forgotten compression techniques such as block truncation/color cells, block VQ and ADPCM-like audio compression

No built-in toilet. 0/10

Hi all, I create hybrid VFX/animations but I'm disappointed with the prospect of GIFs not supporting 60fps animations and the sheer weight of APNGs is also frustrating. I'm also developing a desktop animated image manipulation program called TridentFrame, and with current support of GIFs and APNGs, I'm looking for better alternatives like WebP and JXL to implement, with hopefully achieving good compression ratios and quality retainment to look forward to One of my animations are here: https://discord.com/channels/794206087879852103/808692065076379699/819488572498509844 The cosmic ban blob was a monstrous 25.6MB 700x350 GIF on 50fps

Hi - interesting! Let's take it to <#794206087879852106>

Hello, i'm an image processing / deep learning researcher, with a large background in computer science, and i'm targeting to move our code base to take advantages of new compression format (we have raw uncompressed tif data as input), but we need our data to be losslessly compressed (16bit depth). I'm currently testing out the jxl lib.

Hi, I am a medical imaging software developer with experience in image compression and work for AWS. I am part of the DICOM WG-04 standards body which is exploring adding modern codecs like JPEG-XL to the standard. I have contributed to various open source codecs such as OpenJPEG, CharLS and OpenJPH. My specific interest is around progress lossless viewing codecs and web based viewing. Happy to answer any questions about medical imaging specific use cases and contribute to JPEG-XL

Am a nobody who's really interested in where image formats are going. I think compression algorithms are cool.

Hey, I'm <@456226577798135808> and I hope I won't get nuked again...

What happened? Can't get in your discord account anymore?

Apparently I was sHaRiNg MinOr sExuALiZinG PiCtuReS *cue the mocking SpongeBob meme* Ah, the joy of being in way too many (both SFW and NSFW) servers... Of course they didn't even let me know which server and which photo (every other platform shows you THE offending post/comment) and when I tried to appeal, Clyde (their bot) told me to f**k off without even forwarding that to a single living human! Gonna both re-appeal and send a GDPR request. Unjustified bans are bigger problem, as you can see on their unofficial Reddit and other social media platforms...

sounds epic

Big Discord moment <:Cheems:884736660707901470>

this is my second discord account because the exact same thing happened to me

it happened shortly after I banned some raiders from a server I moderate so I believe I might have been mass reported

at least you have access to the GDPR, in the US you have no legal protection

You could always pretend to be from Cali

Apparently it takes AGES from those wankers to respond to anything, and yes, that includes GDPR requests... if they decide to respond to it at all.

Hi, I'm just a bozo who likes saving MB. I don't know *when* I heard about JPEG XL, but I've probably been your number one closet fan for a while now. Loving lossless compression / many channels / animation / JXL art??. Looking forward to JXL nuking the image formats :).

oof <@456226577798135808>

jxl matrix room when

Did the old one stop working? (It's working)

didn't know it exists

Why people aren't used to search for things first before asking questions.

because searching for this outside actually returned zero results

also i think it bridged only a single channel? and isn't mentioned on jpegxl.info either... yeah idk idk even when it was discussed & i was present, i failed to remember it, i wonder why...

I am a Rust developer working on image processing and color space related projects. I am actively developing my own color conversion lib :3. I am aiming towards implementing jpeg-xl support to the image aspect of my library pretty soon... I also drink lots of coffee and listen to lots of music... idk what else to say atm tbqh

2FA broke (for the third time in my life 💀) So now I am on this account

which methode do you use for 2FA? Don't you get a bunch of OTPs to bypass 2FA in case of an emergency?

I dont have those

I recommend writing down the single use codes somewhere for next time.

¯\_(ツ)_/¯ I do wish I still had said codes

That's why I upload mine to cloud storage <:KekDog:884736660376535040>

I put my 2FA secret and the codes into the same password manager that holds the password to the same service.

Hi! I am a software engineering student about to finish my master's degree. Currently, I'm writing a thesis on modern compression formats such as jxl, avif and webp in the context of medical images. I am currently also writing a simple wrapper for java. Hope I can usefully contribute to libjxl!

Heya!, VulcanSphere (a.k.a. Vulphere) here, just an Indonesian guy who desires a next-generation raster format that can succeed both classic JPEG and PNG and squeeze even more megabytes (I love taking photos and software screenshots). Back in 2018, I discovered the magic of FLIF which progressed into FUIF and naturally I took big interest in JPEG XL (starting with contributing to Wikipedia article of it), now hoping for steady adoption of this promising raster format! Now proudly using .jxl wallpaper in my KDE Plasma desktop. Other interests including watching anime, reading manga, watching motorsport, and listening to music

Yes, that happened. AGAIN. This time for bEiNg a sPaMmEr despite requests to their "support". At least they freed my both e-mails and made them available for registration like if I never had any account here...

Howdy folks, I'm a stereoscopic 3D enthusiast and have many dreams and aspirations, but on the top of my list is to have <#805414133788180491> stop pinging me kthnx

Haha yeah it does get better, at least the distance between levels gets larger

just mute bots pinging u & also the whole channel, it's useless anyway

https://cdn.discordapp.com/emojis/567189169554587658.webp?size=64&quality=lossless

guess I should've blocked it from the get go <:ReeCat:806087208678588437>

whatup ppl

The sky

Hi! I'm just some guy who's into stuff. Special interests include theoretical computer science, cryptography (ask me about homomorphic crypto), machine learning, web stuff - because I'm currently mostly doing (fullstack) web stuff for work - and japanese 2D media (seems like a common interest here). I don't really do much graphics programming these days but I did a software rasterizer (and raytracer but that's easy) for obj files in C++ as an undergrad project and it was a ton of fun (and work). I like JPEG XL because it seems to be the best current candidate for a new standard image format and I'm interested in how features like predictors, patches, splines and dots can be utilized.

Hi, I have already been here for quite some time now but never did an introduction so here we go. I am currently an undergraduate student studying Maths and CS. I got interested in lossless audio compression with FLAC first and then lossy codecs and eventually video (HEVC, AV1, VVC). Image compression is something I've been dabbling with very recently (since the past few months). The idea of something that is otherwise identical to its original counterpart all while having a smaller information footprint intrigues me. (let's ignore that it trades that with computing power for now) Visually lossless but mathematically lossy compression is even more interesting as it is tuned to the human vision and perception, a specialized application. Going forward, I want to keep up with developments in this area and see how far the limits are before eventually the whole image is rendered in real time from scratch xD

im a computer illiterate who's only here because im looking for an alternative image codec/file format to standardize on for my personal website's static and animated images. it's the same reason why im researching the .ogg media format framework for video and audio purposes (av1, opus, flac). i love standardization and maybe .jxl can be a good choice for me.

I am a compression "specialist" taught by many years of experience, I am exclusively a practitioner, every day I compress something, converting one information into another is my fetish, saying that I am a specialist, I do not mean that I am an engineer, I am not even a programmer, in general I am a sysadmin who dropped out of college, my work cor myself is to create a video in a low niche with low quality video or some sd, a couple of days ago a person from the Internet provided me with his xeon and I started compressing anime using my own key, which I standardized for ffmpeg. Due to my experience, I can tell you exactly what to use for any type of information, if you tell me what your needs and goals are, video photo sound text, I know all known algorithms with what known parameters that would definitely advise you this or that, if you need something... The problem is that many users do not understand how it works at all, most people do not even know that you can change the mov mp4 mkv video container without any transcoding of the video stream, as well as cropping the video stream without any recoding. It's very sad, sometimes it annoys me how wrong people compress anything, anywhere in projects. Okay. So, the cost of a pixel for me is much more important than their quantity. I'm also a so-so amateur photographer, and I also know English very basic and communicate only through a translator. And I'm pauper, so I think that behind the high, absurdly high, uselessly high qualities, everyone has "Got greedy." After all, the higher the quality, the better. The more pixels the better. But is not. I don't like it. Youtube is one of the main hotbeds of this problem, too low bit rates per size.

Thank you for the warm welcome! I'm interested in minimal, energy-efficient computing — from a degrowth/climate/conviviality perspective. (e.g. I have shamelessly spent hours investigating dithering and color reduction in comparison to other compression approaches). I am also into tool building and makings things simpler for people online, and for all those reasons, JPEG XL came up as an awesome project!

Dithering and color reduction? Just like here? https://solar.lowtechmagazine.com

That is exactly where it started! Though eventually, I came across this (https://www.simplethread.com/why-your-website-should-not-use-dithered-images/), and then this (https://doodad.dev/dither-me-this/), and then finally https://squoosh.app/ — where, with a lot of experimentation, I realized the biggest gains in compression are mainly through different format, color reduction, and lowering quality! (Not so much dithering.) Anyway, there's my unabridged introduction about how I came down the rabbit hole of image compression and got so excited to find jxl!

Hello everyone! I am a research scientist at a biometrics company. We have a quite interesting use-case for JPEG XL and was hoping if anyone can point me in the right direction. Basically, we want to build a customized version of JPEG XL specifically for compressing human faces ( passport pictures). The goal is to go below 1KB and still maintain enough fidelity for humans to compare it with live on a mobile phone screen. Just wanted to get ideas on the feasibility and the efforts behind this. Thanks!!

What kind of resolution would be needed? I can get down to 2KB pretty easily with a 128x128 image.

Hi <@176735987907559425> . So we are looking at a resolution of 120x160 (widthxheight) pixels at minimum or double of this at max. The major point is that the image size should be around 1KB (hopefully less).

let's take this to <#794206087879852106>

article ded

seems to work on my side... ig it's back up

Hello everyone!. I ran JPEG XL on linux. However, I want to use JPEG XL on android and IOS with C++. How can I do that? Can you share an example? Thanks!

With c++, you can use libjxl. Here is an example of loading an image on android : https://github.com/oupson/jxlviewer/blob/2d70c1b14e718ee43c46a312367552d0e37e5a6d/libjxl/src/main/cpp/native-lib.cpp#L48

Thank you so much! I will try it.

Is there any example for IOS about JPEG XL

Is it suitable for IOS : https://github.com/yllan/JXLook

it kind of works but it's an old version of libjxl and it has only a primitive user interface

Can you share an example about iOS?

Oh iOS, no idea, I don't have iOS devices, only a macbook

How can I do regional image compression with JPEG XL? For example, in a portrait photo of a person, I want to compress the face less and the background more.

How can I play with the DCT coefficients? I want less compression in the center of the image and more compression outward.

Maybe better to discuss this in <#804324493420920833> — I don't think there currently is an easy way to do this but it should be possible in principle (and perhaps it would make sense to expose something like that in the libjxl encode api)

Hello ! Should I introduce myself ?! Some years ago, I discovered FLIF, played with it a bit and was amazed by the concept and pleased by the mindset of its author... which I followed when he jumped on the jpegXL train. I planned to be just another user until chromium did what it did. Now I would be happy to help making this format the next jpeg extra large 😄

EXTRA BIGGER

Greetings! I'm just a freelance illustrator and hobbyist photographer who tries to save some disk space. 🥹 Also sometimes code during my free time~

How dare you want to save disk space. AVIF should be good enough for everyone.

So saith the nameless "leadership."

Nameless and therefore blameless

Oh, you're the one who wrote about the photon noise JPEG XL thread, right?

Twitter? Yes I did

which thread ? 🙂

Hello everyone! 🙂 I work for Solaris Synchrotron at Jagiellonian University and I am involved in LEAPS-INNOV WP7 group on data reduction. I am not a data reduction specialist, rather a simple guy who knows something about relational databases and data compression overall. I'm one of the guys tasked to research useful compression formats that could be implemented either in hardware (fpga or cpld) or in Blosc2+HDF5. Such scientific facilities experience large influx of data required to store, while numeric values can be compressed in various ways, images from detectors are really tough to compress and they require lossless compression almost all the time. You don't have to convince me to JPEG-XL as I'm already convinced, but I would like to get some help on influencing our group that this compression format can be sufficient to store large amount of scientific data. Three things are important: great compression ratio, speed (It must be quick enough, meanwhile cjxl's lossless compression is slow as hell) ready to implement in Blosc2 or HDF5.

cjxl's lossless compression is by default a lot slower than cjxl's lossy compression is by default, but that is configurable. I think the slow lossless default speed is because they want lossless mode to be significantly smaller than other formats by default. But lossless encode speed is very very configurable. The smaller EFFORT settings are very very fast, especially `-e 1` mode is insanely fast.

thank you, I will try it out and make a table comparing different effort values and results

the main attractive thing in jxl imo for scientific applications is that you can have arbitrary bit depth and arbitrary number of channels and it will compress very fast and quite effective at e1

One thing to note is that the upcoming, unreleased, development version (which will probably become 0.8) has a much faster `-e 1` mode that is based off the fast-lossless encoder. The latest release version, 0.7, does not have that yet. But you still have a lot of different speed settings to choose from. The fast-lossless encoder has absolutely insane gigapixel-speeds thanks to veluca

if your sensors have a precision of 13-bit or whatever, that's fine for jxl — while most other formats (png, tiff, exr etc) require bit depth to be a multiple of 8 and if it's not, you'll basically have to add padding bits and pay for them

I know nothing of Blosc2/HDF5, but implementing an equivalent of fjxl in hardware *should* be feasible (well, not that I know much of vhdl and friends, either) - but even in software, if you have a modern CPU you can easily encode 300MP/s on a single thread

Yeah, the fast-lossless encoder is really great, the code is in the fast-lossless folder in the libjxl source code, and it's ludicrous speed.

Also it parallelizes nearly perfectly, so with 8 threads you can get 2000 MP/s or so.

Well, it parallelizes to your memory bandwidth, pretty much

does more threads = worse compression though?

No, makes no difference at all for the file you get in the end. Not at e1, not at any other effort either.

(as opposed to avifenc where more threads means worse compression)

That's pretty nice and pretty uncommon

I... I think I never introduced myself <:KekDog:805390049033191445> So I'm currently an engineer student, almost finished my studies. And... I'm really a big geek when it comes to compression <:CatSmile:805382488293244929> I can read hours and hours of documentation about compression codecs, their efficiency, ... But most of the time I program stuffs from myself, and try to share them whenever I can. Stuffs I love: developping algorithms, do optimizations and really low-level programming.

Hey all, I'm Simulping. Currently a freshman at senior high and been interested in AV1 and codecs in general since 9th grade. Started from a weird obsession about uploading entire movies to discord.. Currently adjusting myself to Linux and doing the great deep dive into how this entire thing works (codecs included <:dogelol:867794291652558888>)

I never introduced myself! Hello, I’m Gianni & I am a freshman at WPI. I co-develop Aviator on Flathub for AV1 encoding, and I like all things codec related. I am currently in the process of starting a nonprofit organization dedicated to free hosting solutions for FOSS projects called the Radix Project, and struggling with the legal process & paperwork. My biggest achievement is getting to #3 on hacker news for my JXL article. I am a big Jon Sneyers fan <:PepeOK:805388754545934396>

Hi, been interested in JPEG XL for a while

Hi! I make black and white mathematical images, so I'm interested in how JXL compares to 1-bit grayscale PNG

JXL should be more dense in general. Passing `--override_bitdepth=1` to cjxl may yield a size improvement over default lossless.

You might find further improvement were you can directly represent your algorithm as an MA tree. <https://jxl-art.surma.technology/wtf.html>

?

Hi, I'm Dzuk, I create and maintain an alternative emoji set called Mutant Standard (https://mutant.tech/), as well as a companion tool called Orxporter (https://github.com/mutantstandard/orxporter), which allows people to batch export custom SVG emoji sets. As someone who maintains and distributes a very large emoji set, lossless image compression is pretty important to me and the people who use it, and thats how I discovered FLIF and then JXL ^^

👋 Hello! I'm here cause I'm interested in seeing the future of JXL.

random Linux user + artist curious how this works https://tenor.com/view/homestuck-nepeta-cute-gif-5666407

Hi! I am Cacodemon345. Author/maintainer of: 1. UEFI-DOOM (forked). 2. Blast Thru Reborn (attempt to recreate the Arkanoid clone of the 90s game, Blast Thru, currently on hold). 3. DoomGeneric NTNative (Doom running at Windows XP's chkdsk environment). 4. DoomGeneric NTDrv (Doom ported to run in Windows XP's kernel mode). I also contributed to the x86 emulator project 86Box, and also to GZDoom.

Welcome! What an achievement!

Hi, I'm axos. I'm a dev working mostly on web apps. I've recently gotten into image processing in order to build an open source software to optimise all types of files before backup (family takes a lot of photos and videos).

Welcome! Jpeg xl is what you need 😁 especially if you have mostly real photos

It's in my todos as well but it's gonna be the end user's decision conversion if he wants to use it. I'm planning to release the code after I finish png and mp4 (if it's not too hard) and then take a pause to learn more about each format. Currently I'm only using libraries and I'm not that confident on letting other people use it before backup or recommending a specific format.

The good point is that, with the right encoding options, you can revert to your original jpeg, from jpegxl

Interesting projects there!

Hi, I'm retro. I've started reading about JPEG-XL recently (took up an image compression course at my uni so kind of a newbie here) and it seems pretty interesting ngl. I wanted to know more about how the meta adaptive context modelling works (MAANS). Is there like a documentation about it or if someone could tell me where do I start reading about it ? TIA

unfortunately I am not sure there's much better than the source code ... the tree learning happens here https://github.com/libjxl/libjxl/blob/main/lib/jxl/modular/encoding/enc_ma.cc and the core decoding happens here https://github.com/libjxl/libjxl/blob/main/lib/jxl/modular/encoding/encoding.cc#L130 although speed optimizations make it less clear

Hi, I'm Ayman. I'm always willing to learn something new. I'm working on my PhD in image compression. I have implemented before arithmetic coder, ANS coder, legacy JPEG codec, and pseudo-JPEG2000 codec. I'm currently interested in practically fast lossless image compression, and I was impressed by the performance of JPEG XL. My Github is github.com/a1024

I'm Zach, I'm not that good at coding but I have a very big interest in technology, and have had that interest since I was 7 years old. When I do code, it's usually very janky and broken lmao, Still don't know how I pulled off a [header parser](https://github.com/DexrnZacAttack/DexrnZacAttack.github.io/blob/main/js/QMGParser.js) in JS. My [main project](https://github.com/dexrnzacattack/QmageDecoder) ATM is trying to decode an undocumented image format (QMG) that Samsung uses for Boot Animations and assets in APKs. My website is https://dexrnzacattack.github.io

threads > 1 means worse compression, because it enables row-mt, any higher makes no difference

hello, i'm just a artist interested about jpeg xl. i dont do any coding whatsover (i'm planning to change this very soon) you can find my artworks on my twitter : https://twitter.com/eunaoseibrother

im only here because i have to compile libjpgxl all the time and i saw a discord invite in README.md hi

who put that there

Hey, I'm Red 👋 . I'm currently working on a high performance multi-target (cpu, spirv and vhdl) implementation of JPEG XS. Since XS and XL are both newer JPEG codecs I thought I'd join this server. I'm no expert on image/video compression by any means but I'm quite experienced in C++ and hardware design.

Hi Red — are you with IntoPix or Fraunhofer or not related to the main proponents of JXS?

Hey \_wb\_, I'm not related with any of the entities that are involved in the standardisation and licencing of JPEG XL. I'm part of a university work group and am working on the implementation together with another co-worker. We are doing this as part of an academic project and are looking at potentially making our implementation open.

That would be cool, it would still be patent-emcumbered though, I guess

I can bring you in contact with Tim if you want

Yes unfortunately, however it would be accessible for educational and non-commercial applications I believe.

Or Siegfried. The JXS chairs at JPEG.

They would be very happy with an independent implementation I assume

Let's take this to <#805176455658733570>

Hi, I joined this server to extend my media processing related server list ||Oh, and I wanted to see Fab again <:PepeHands:808829977608323112> ||

we share 11 servers now 💀 10 of them encoding related

Hello everyone I've been in the subreddit for a while and I just found this server. My first post is about Jpegli however

What other encoding related servers are there? (Feel free to dm)

Hi all, I've reached out to <@794205442175402004> before, but I've meaning to introduce myself to this group. My background is on game development and GPU hardware architecture. I'm currently developing [Spark](https://ludicon.com/spark/) a real-time GPU texture compression library. One of the most interesting use cases of our codec is to allow games and other real-time applications to use formats like JXL to distribute texture assets, but still enjoy the benefits of GPU texture compression. In my experience this is a more interesting approach than Basis and other RDO codecs, especially when targeting low bpps. My most immediate goal is to familiarize myself with JXL in order to help guide our clients that are attempting to go this route.

Hello. I just translated j40 to evaluate it and am only slightly interested in JXL. I once worked in video coding. Created: gamut an image library for ease of use in the D language. I made a toy codec once, QOIX.

Ironically I got to know QOI by the article from <@794205442175402004>

QOIX is inspired by QOI (like many others) = 3 different QOI-like codecs, one for rgb8/rgba8, one for 1 or 2 channels, one for 10-bit. It's anecdotic. It works best to replace 16-bit PNG if you only need 10-bit (damn, the mesage rate limiter on this server is annoying)

QOI is the Toki Pona of image formats to me so I didn't expect much from it

it fills the "fork your own codec" use case ^^

to avoid the message rate limit, go to other channels like <#805176455658733570> or <#794206087879852106> 🙂

I like this analogy

so which one's Lojban?

and what is tiff in this analogy !?

tiff is Church Latin

so, regular Latin, but pronounced by Italians? 😛

> Ecclesiastical Latin, also called Church Latin or Liturgical Latin, is a form of Latin developed to discuss Christian thought in Late antiquity and used in Christian liturgy, theology, and church administration to the present day, especially in the Catholic Church. It includes words from Vulgar Latin and Classical Latin (as well as Greek and Hebrew) re-purposed with Christian meaning.[3] It is less stylized and rigid in form than Classical Latin, sharing vocabulary, forms, and syntax, while at the same time incorporating informal elements which had always been with the language but which were excluded by the literary authors of Classical Latin.[4] > Its pronunciation was partly standardized in the late 8th century during the Carolingian Renaissance as part of Charlemagne's educational reforms, and this new letter-by-letter pronunciation, used in France and England, was adopted in Iberia and Italy a couple of centuries afterwards.[5] As time passed, pronunciation diverged depending on the local vernacular language, giving rise to even highly divergent forms such as the traditional English pronunciation of Latin, which has now been largely abandoned for reading Latin texts. Within the Catholic Church and in certain Protestant churches, such as the Anglican Church, a pronunciation based on modern Italian phonology, known as Italianate Latin, has become common since the late 19th century.

basically a hairy mess of a language with diverging subforms and poor standardization, and now not really used anymore but niche usage remains

that's TIFF for me 🙂

It's just classical or medieval latin but with niche-specific jargon and not very medieval anymore since the official latinists at the vatican try to keep a very high standard of classicalness.

Hey, have been skulking around here for ages, but thought I'd officially say 'hi'. I'm from the stock photography world – ages ago I ran iStockphoto and later R&D at 500px, so have been interested in image compression for decades. Have a comp sci degree, but mostly do marketing and product work now. And I have a bit of a proposal for <@794205442175402004>… Also years ago wrote articles like this: https://iso.500px.com/500px-color-profiles-file-formats-and-you/

Cool! Welcome here!

I recently came here after learning about JPEG XL from a Gwern article about unexpected turing completeness. I saw the JXL art page and thought it was pretty neat you could make cool images like that with such a small file size. I'm curious if it would be possible to make a very small sized Mandelbrot/Julia image or if that is just not how JPEG XL fundamentally works

do you have a link to the article ? jpegxl only partially works like that 😄 it uses more than one strategy to encode, MA trees are one of them, have a look at <#824000991891554375>

Here is the link: https://gwern.net/turing-complete And neat, I'll check that channel out. Any of type of self repeating images I can make would be primo

Mandelbrot/Julia is unlikely to be possible within just an MA tree but there might be some clever way to do it that nobody hasn't thought about. Sierpinski is quite easy to do, and also the XOR pattern can be done, see <#824000991891554375>

Hello everyone! I created https://github.com/olokelo/jxlpy and this Pillow PR https://github.com/python-pillow/Pillow/pull/7848 (isn't merged yet). Hopefully brings more jxl into Python's world 🙂

Hi. Been in this Discord server for some while now, but never got to introduce myself yet. Time to rectify that. I'm Lumière Élevé, who has some interest in media-related projects, but experiences with media codecs are amateurish or among the above-average users at best. Designed some toy lossless image format before, which divides everything into 16 by 16 grids and fill for mismatched pixels. Currently I'm mostly known for MIDI-related projects like Octavia and a few others. Has been an advocate of JPEG XL ever since discovering it back in 2023, but never got to join the server due to broken invite links, until I made a relatively large test a while ago and tried to join again.

Welcome! Are you a relative of "Lumière à tous les étages"?

~~nope~~

Hello. Here from the link posted by [Theo](https://x.com/t3dotgg/status/1829361633614840168)

Welcome!

It was widely discussed at <#822105409312653333>

oh, neat!

Hi everyone 🙂 I'm Access, JPEG XL nerd and software engineer working in Rust ~~This is basically my entire personality so nice to meet you~~

Hello everyone! I'm a master student in visual computing and quite new to codecs at a low level. I'm currently exploring ways of how variable rate compression algorithms could be used in real time rendering to trade a bit of efficiency for more texture storage space. I really like jxl for the multiple channels it supports and the groups that can be decoded independently/progressive decoding.

I notice a lot of new faces

Hi everybody, I’m here because I would like to experiment with some of the more interesting features of the format (to me at least): 1. Patches and 2. Zero-duration frames. The spline stuff is fascinating too. I’m generally interested in trying to understand a format by looking at (or hand creating) minimal examples of the format in hex editor. JXL fascinates me because it combines two interests: 1. Photography and archiving photos for future generations, and 2. Book scanning and formats for archiving and/or presenting the pages.

hi what is this jpeg xl thing that everyone is talking about, is it like avif or webp?

Almost, except this one is extra lewd

Hi. I'm a CS student with an unrelated focus. I was led to JPEG XL by a desire for lossless images with at least the same metadata capabilities as JPEG. Accidental data loss pains me quite a bit, so I'm looking to JPEG XL as a possible solution for the preservation of personally important images and their metadata. ("Accidental data loss" referring to metadata removal and/or lossy compression performed during the transfer of images through various channels, such as sending over Discord, e-mail, or SMS, uploading to cloud storage such as Google Drive, and/or any other context under which an image might lose information without my explicit intention being to do so. I'm not sure of every context under which this does or does not occur, but my hope is that JPEG XL will help reduce the situations under which it does occur.) I'm hoping I can learn enough to contribute and further adoption efforts, but don't expect I'll be able to do so quickly.

> metadata removal and/or lossy compression performed during the transfer One benefit of JPEG XL in that regard is that there is not as much software knowing that a .jxl file is, so they wont touch it

Good point. I should HAMPER adoption efforts as much as possible.

average web browser:

You can also just plop the file in a ZIP

listen

hi im just a nerd and i think jxl is cool

JXL is cool

See <#794206087879852106>

Hello everyone, I am new to this forum, and first time on Discord as well. I am a hobbyist programmer, and was looking for image formats that supported high dynamic range and lossy compression, for an application I am working on, when I read about jxl. I have downloaded the library, and after some fiddling, managed to build it from Visual Studio, as a static library. I have used the test-applications encode_oneshot.cc, decode_oneshot.cc and decode_exif_metadata.cc as the starting point for a wrapper VariantImage struct that I am using as my interface to the library. I am really impressed with JXL, the compression is amazing! Thank you to the developers, for this wonderful image format!

Glad what you found isn't JXR

Or AVIF

Someone invited me here, i know nothing about jxl, but i love that it is real, more places would allow such yummyness

Apparently I fell asleep on my phone and sent them a few invites, but they showed me [Nanopixel](https://www.escapemotions.com/blog/prepare-for-perfect-prints-with-nanopixel-technology), which makes for some extremely challenging images to compress

very yummy indeed

```ansi AI Engineer | GPT-4 | Agents | LangChain | RAG | Web3 AI | Vector DBs | TTS/STT | Fine-Tuning``` Hey everyone! I’m an AI Engineer building intelligent, scalable systems at the intersection of deep learning and modern tech stacks: 🧠 LLMs & Foundation Models (GPT-4, Claude, Mistral, LLaMA, Gemini) 🔄 RAG Systems, Embeddings, Vector Search (FAISS, Pinecone, Weaviate) 🤖 LangChain, CrewAI, OpenAgents, AutoGen, ReAct, Toolformer APIs 🧑‍💻 AI + Web3 Integrations: Smart Agents, NFT AI, Token-Gated AI Tools 📊 Advanced NLP, Prompt Engineering, Retrieval Pipelines, Memory Graphs 🗣️ Voice AI: TTS (xtts, Bark, ElevenLabs), STT (Whisper, Deepgram) 🧪 Fine-Tuning, LoRA, QLoRA, Quantization (GGUF, GPTQ) 🌐 Backend APIs, Automation, DevOps (Docker, Supabase, Node.js) Whether you're building AI agents, intelligent dApps, or SaaS tools powered by LLMs, I bring battle-tested experience and fast delivery.

OpenAI has retired GPT-4

So, what brings you here?

it is the kind of person a random business man with no technical knowledge would bring in the company when no one asked x)

welcome! we talk about plenty of topics that you seem to have no interest in whatsoever. feel free to look around and if you don't understand what you see you can always ask chatgpt 🙂

you just missed the IPO of the JPEG XL token, but if you need someone to recompress monke pngs we can help -# really hope this is a spambot and not a real person that I just bullied

Argh dang it! I wanted to mint 20 JPEG XL NFTs! I thought I vibe coded a blockchain monitor to send me a notification on my Neuralink!

Hi 🙂 Looking to learn about Jxl to mint some smaller file NFTs on chain

Look into <#824000991891554375> , very small files, some even shorter than a CDN URL when base64 encoded . Could easily put them on chain

I think I did not introduce myself yet. I was a researcher in computer graphics and moved to image processing recently. I'm in R&D, managing image processing for astrophotography. JPG is the format used for output images, but FITS is where the raw data is. We're on the lookout for an intermediate format that can be used in the middle of the chain, with controllable perceptual loss. JXL sounds perfect for that, but it's not quite suited to images that have a linear dynamic, similar to raw output of cameras. I'm trying to work out all of these issues as they appear. In any case, I love the format, I think it's brilliant, and I think we all deserve better support for it.

I think I did not introduce myself yet.

Oh yeah, haven’t introduced myself here, my scope is probably not as big as other ppl here <:PirateCat:992960743169347644> So hello everyone, I’m currently collecting various mangas for my own reading and is in the process of converting all of my digital manga collection to use JXL to reduce file size, because I can’t trust webp or avif to not shittify the images <:kekw:808717074305122316>

Are the mangas coloured or greyscale?

A mix of both, though most of it is greyscale

It's said JXL isn't very optimised for greyscale non-photographic images now

Hi! I'm a math student and also interested in all things computer-ish. I am making my own image library for fun (it's nothing fancy at all, currently just a way for my other programs to create an image pixel-by-pixel and then save it in a sensible format) and I just decided I'd add support for exporting to JXL, because why not. Also I do some photography and when (I hope) the format is more widely adopted I would love to be able to share images with better quality for their size (e.g. on my crappy website).

Hi, its meeee and im good at mabye doing cool things somtimes