JPEG XL

How does Splines related to predictors? It looks Splines are added after predictors three? Is is possible to combine NotLast and Splines somehow? It looks Splines can't write on Alpha?

Talking about Splines, looks like Firefox does not support it yet?

Splines only write on RGB (or XYB/YCbCr), and they are additive (get added numerically to whatever pixel values are already there)

Firefox should support them but there might be bugs in the libjxl version they use

There's also a limit on how many splines you can have (and how much pixel area needs to get drawn)

This document wg1m97061-ICQ-Study_text_of_CD_of_18181-1_2nd_Edition.pdf has good information that completes the one find in jxl-art.surma.technology help page

It looks like FramePos does not work with Upsample or may I just dont understand it 🙂

yeah there's a bug in jxl_from_tree there iirc

jxl_from_tree is a bit of a hack and some combinations don't work

one thing that is missing from it is a way to set the dimensions of a frame (which can be different from the image dimensions but currently they're just always the same as the image dimensions)

Curious question, why these operators and transformations are "so limited". I'm not into the compression area, I believe this may be related how predictors are supposed to work and having hardware encoding/decoding implementation in mind?

the main limitation is that you want to decode in scanline order and can only use already-decoded pixel values in predictors / decision nodes

scanline order, I understand...but a bit more complex "ISA (Instruction set architecture)" could have been more useful, at least, for "jxl-art" proposes 😛

it's not designed to make art, it's designed to be useful to make image-specific context models for entropy coding 🙂

question regarding share on jxl-art.surma.technology, will that be somehow made public ? How long will it last the URL? I'm preparing my entry for Inercia2022, I don't want it to be public yet but I want to share the URL on the readme file

The url contains the source, there is no server-side storage of anything

But you can also just put the source itself into the readme so people can use it with jxl_from_tree even if the web interface would somehow no longer be available

yes, that's it 🙂

do you have already 3 entries for the compo? hope so! 😛

jxl-art.surma.technology: funny I just noticed that the png export actually exports the alpha channel and it is useful for debug. it may be a feature to visualize the alpha channel on the browser

which leads me to the question: what is the default background color rendered by jpeg-xl ?

I suppose the answer is: it renders a transparent color ? :S

what do you mean by a transparent color? JPEG XL supports alpha channels and alpha blending in frames as well, but if your frame doesn't have alpha channels it is fully opaque.

I have alpha channel

then it will get rendered, unless jxl-art.surma.technology did otherwise (I'm less sure about that)

yes it is all ok, my question does not make sense. I was asking "what is the default color of a transparent (final) image" but it depends where it is rendered / software / color theme (eg webpage)

There is no notion of a default background color in jxl - if a pixel is transparent, it's just transparent

I was looking if there was a way that I can fully "clamp" set the alpha channel to one value I have multiple alpha channels (that can has negative or unknown values), I was looking if I can set the end result to a value, but I can't find a way with the option on help page

PNG does have such a notion, but I don't think any software cares about it

Multiple alpha channels? You're not using jxl_from_tree or the web app then?

sorry I mean, a composition of multiple NotLast it is just one alpha channel in the end

Ah I see

I actually kinda wanted to make some graphical interface to jxl_from_tree (or even my own encoder), preferrably with something like Blockly

Ooh that would be cool

There are more bitstream features to be exposed too, as you will know very well having implemented most of them 🙂

E.g. setting custom upsampling or gaborish weights

yeah, and that directly relates to testing

the only difference is whether the user input is textual or graphical (and visual languages like Blockly remove the boundary between two)

(and being able to represent a faulty file, but this is a separate concern)

so no suggestion how to clamp a channel value? the decision nodes have clamp values to evaluate but leaf nodes dont have :/

I must be confused with something, since there is the Set leaf node and the way the "NotLast" thing works on relation to the alpha channel

There are various blend modes, default is regular alpha blending

Set sets the value in the current frame (layer) but what the value is after blending cannot be set

I couldn't fix the alpha issue. I hope that it will be displayed ok on a clear background (without checkerboard effect)

My entry: png screenshot size: 912kb, readme size: 1.1kb, jxl size: 71 bytes 😅

You can always just add a first layer that has alpha set to 255 and the background color to whatever you want, no?

I tried but it didn't worked for me.. I was not sure what "last" is with Alpha enabled (didnt' investigated much time) but I couldn't ended with some expected result ``` Alpha Bitdepth 8 NotLast if c > -1 <-- Is this the last? - Set 255 - Set 255 NotLast ... other things ... if c > -1 <-- Or is this the last? - Set 255 - Set 255 ```

Last refers to frames themselves each frame contains channels, like possibly an alpha channel

the Last frame is just that, the last frame

as far as I'm aware surma's website generates frames in the order specified

Hello group, I am trying to understand how to generate art. Starting from the basic. ```if c > 0 - Set 0 - Set 255``` This generates a red square. Why red? and how would I choose other colors. I have found others by trial and error but want to understand how is it working. I have read https://jxl-art.surma.technology/wtf.html but still not 100% clear. Thanks!

by default, the channels are R,G,B

which are indexed by `c`, starting from 0, so Red is `c==0`, Green is `c==1`, Blue is `c==2`

```if c > 1 - Set 255 - Set 0``` This gives me blue, shouldn't be green?

the first branch will be taken when c>1 which is true for the B channel

because blue is 2?

yes

so how do I get green? since ```c > 0``` gives red and ```c > 1``` gives blue. ```c > 2``` goes back to white. I assume because 0,1,2 don't pass ```c > 2```

Found it! ```if c > 0 if c > 1 - Set 0 - Set 255 - Set 0 ```

Now to understand why

it's just a bunch of if-then-elses

if (c > 0) { if (c > 1) { val = 0; } else { val = 255; } } else { val = 0; }

does it always start on 0? so in this example the first if does not work at first right?

I think I understand it now. That if has RGB so this is yellow ```if c > 0 if c > 1 - Set 0 # BLUE - Set 255 # GREEN - Set 255 # RED```

Thanks for the help!

you can also work in other colorspaces than RGB, often it is nicer to work with luma and chroma than with RGB

e.g. using `RCT 1`

(which is YCoCg)

Thanks i'll explore those

Italy flag (Green, White, Red)explained: ``` if c > 0 if c > 1 if x > 340 if x > 680 - Set 0 # We want red at the end, so (R:255 G:0 B:0) This sets blue to 0 for the end part. - Set 255 # If not the end then the middle of the flag we want white so (R:255 G:255 B:255) This sets blue to 255 - Set 0 # if we at the beginning we want green (R:0 G:128 B:0) I choose a not so bright green that's why 128. This sets blue to 0. if x > 340 if x > 680 - Set 0 # We want red at the end, so (R:255 G:0 B:0) This sets green to 0 for the end part. - Set 255 # If not the end then the middle of the flag we want white so (R:255 G:255 B:255) This sets green to 255 - Set 128 # if we at the beginning we want green (R:0 G:128 B:0) This sets green to 0. if x > 340 - Set 255 # Red is 255 at the end and in the middle this sets both. - Set 0 # If at the beginning we want red to be 0. - Set 0```

Deadline is getting near! Don't forget to submit!

**0b5vr** _“Iceberg”_ 2022 image/jxl 34 bytes https://jxl-art.surma.technology/?zcode=C3IOUTCy5AotKE7MLchJVTDicsosSUktKMlQMDTg4spMU0hWsFMw4FJQADIDilLLgDxDI1MgX0FBV8HPVUHXCMoOVzCEqAqHagAJOpal%2B2n7hSvoGkIFglNLFEyMLbgA

<@541014985627926551> yours slaps!!!! wtf

Amazing, is the deadline already opened?

Impressive entries!

the compo is finished rn!

<@232233105040211969> you found a nice set on such a few bytes

Will the entries be put online somewhere?

I think they will be on https://files.scene.org sometime soon

**byteobserver** _“Block Nebula”_ 2022 image/jxl 98 bytes https://jxl-art.surma.technology/?zcode=xVLBDoIwDL3vK%2FoDS9iERC4mmhA5LSQcpldlCkcRjTF8vGND7MAMb%2B7UvbbvvTbNLzelnors9hsiq6IpgcU8IKmqzmUDLOAhIdUJjrCCkADoUOow0KH5tLLVX1MG5lHIVdPnbTkVn44eE1QkDmbQR8cUcwS%2Bi8Wo2ApJ41EVwEYp3ZRltbqvD9ekrr%2F0Ot2UebMRmzjqlsGiqaikUs6pTcV%2BcDvn1%2B%2FYt8X%2Feut2tk0ntA4hX5JZu%2FjobGzbhgPFqeF2XWnPqEgT%2B8c2FxzzaBoPhzMTnkfDLw%3D%3D

**byteobserver** _“Something I didn't end up submitting to Inercia”_ 2022 image/jxl 56 bytes https://jxl-art.surma.technology/?zcode=Cy4sTU2tSuVyyixJSS0oyVAwNOMKcg5RMObiykxTCFewUzDgUlCAMg0NjIyBPAUFXYXg1BKwDERO1w%2BmEirip%2BvniiQCVRUejiIGMic8NTM9oyQ1RcEQRdi9KDElMzWvREHXzNTU2IxcY%2BDCuoa0cJsh8UboGuI2GhKaxhYmXAA%3D

https://www.youtube.com/watch?v=X1T7RUP9SpY&t=2438s

here is the one I submitted

source code requires a recent jxl_from_tree, won't work on the current web version

``` /* Background layer: Weighted predictor cloudy stuff with some noise added */ Width 1024 Height 576 Upsample 2 Bitdepth 9 Alpha NotLast Noise 0.1 0.05 0.05 0.05 0.05 0.05 0.05 0.0 RCT 13 if c > 2 - Set 511 if c > 1 if y > 0 - NW 1 - N -1 if y > 432 if WGH > 0 - AvgN+NE -3 - AvgN+NW -1 if c > 0 if y > 0 - N 2 - Set -500 if y > 30 if N > 300 - AvgW+NW -300 if N > -50 if WGH > -3 - AvgN+NE 7 - AvgN+NW -3 - AvgN+NW 100 if W > -200 - Weighted -2 - Select 350 /* Overlay 1: a spline of light that will go around the JXL logo */ FramePos 512 164 Upsample 1 BlendMode kAdd NotLast Spline 0 -0.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2.0 1.0 -0.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 20 48 992 128 128 196 880 256 220 320 832 384 128 448 920 512 64 544 EndSpline RCT 0 - Set 0 /* Overlay 2: Sierpinski-ish triangles to go below the JXL logo */ FramePos 512 640 Upsample 1 BlendMode kBlend NotLast RCT 0 if c > 2 if y > 0 if N > 0 if NW-N > -1 if N-NE > 0 - Set 0 - N + 0 if N-NE > 0 - N + 0 - Set 0 if NW-N > 0 if N-NE > -1 - NW + 0 - Set 0 if N-NE > -1 - Set 0 - Set 280 if x > 511 - Set 1 - Set 0 - W 1 ```

``` /* Overlay 3: a JXL logo */ FramePos 512 164 Upsample 1 GroupShift 3 BlendMode kBlend NotLast RCT 13 if c > 2 if y > 0 if y > 336 if x > 510 if x > 511 - NW 0 if y > 383 - Set 0 - N -10 - NE 0 if x > 511 - NE + 0 - NW + 0 if x > 1022 - Set 0 if x > 928 - W 10 if x > 927 - Set -450 if x > 895 - Set 0 if x > 801 - W 10 if x > 800 - Set -450 if x > 223 - Set 0 if x > 128 - W -10 if x > 127 - Set 500 if x > 95 - Set 0 if x > 1 - W -10 if x > 0 - Set 500 - Set 0 if c > 0 if x > 0 - W 1 - W -1 if y > 0 - N 3 - Set -560 /* Overlay 4: a signature (JS) in the bottom right */ FramePos 1860 950 BlendMode kAdd Upsample 1 Spline 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1.0 0.3 -0.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 40 75 55 70 55 160 10 150 70 130 110 120 40 100 120 80 EndSpline RCT 0 - Set 0 ```

**\_wb\_** _“something I didn't end up submitting to Inércia”_ 2022 image/jxl 50 bytes https://jxl-art.surma.technology/?zcode=VY5BC4JAFITv71fMXYS1ti0vQUXRaZMs9iz6qgUtsSX6%2Ba1loreZN%2FM9Zm1dwbW7IRJkbOFFrATt2V5vDjMp6Fw%2Fs6ouGZIOjeW7y5x93BHRcXPCJKZdk1WcPJ4QEET2ghxLnwJevr0UXgIhDMKol12e97k3ScOvFo1%2BtbaYst8gRO9Xr6sJNMLFn9H9g681LT%2BZd4eO0IHeIhwetd8yHVNSymFh3E%2B55NxBKUUf

ooo, this looks like a fantastic test of features

beautiful

https://embed.moe/https://cdn.discordapp.com/attachments/824000991891554375/1038836511694667897/wb_-JXL.jxl

<:thenk:349311010584395788> I didn't change anything

Art is imo always caused by creativeness sparked by the constraints of its medium. I think jxl-art is the first time the medium is an image compression bitstream, though you could argue that the concept of pixel art (where the 'bitstream' is simply some uncompressed image format, and the pixel dimensions being small is what compresses/constrains it) and also gpu shaders are older examples of this.

btw, just want to double check, is this submission CC0?

and/or public domain

most of your jxl art is by default iirc but idk if the contest has some rules on that

In a way, the constraints of, say, a sonnet (14 lines, iambic pentameter, rhymes) are not that different from the constraints of conforming to a particular bitstream syntax. It's a set of ultimately self-imposed (or culturally standardized) rules, and I believe such rules and limitations, even if they are arbitrary, help to trigger the problem-solving creative areas of the brain that paradoxically make it easier to achieve artistic expression than if there are no rules at all — a blank piece of paper or an empty canvas is more intimidating and creativity-crushing than a situation where one thing leads to another because the constraints need to be satisfied.

Yeah it's CC0, it has my signature on it already so no need for CC-BY or something like that

I don't know about the other submissions and if the contest had rules about that, best to ask the respective authors

jxl files are archived to scene.org file storage but I believe each authors have the right I believe. I explicitly grant you (_wb_) a right to use my work without my perm (with credit preferred) if you want to use this for jxl advertisation purposes

It's a good idea to discuss licenses. I will put all of my works previously posted in this channel (as byteobserver or nwoeanhinnogaehr) as CC-BY

https://twitter.com/jonsneyers/status/1589707844718579712?s=20

<@232233105040211969> congrats with the new role 🙂

**\_wb\_** _“The Magic Tree of Healing”_ 2022 image/jxl 39 bytes https://jxl-art.surma.technology/?zcode=VY3BCoJAGITv%2F1PMPRZ2yw26BCahJ4MsPEuuumC62Bb19v1GLHX7vmGGKW3tO2zWkjJj285DR5J21tfGca4kpdN4d0VnG48VHZMTNB0mawZfeTsObGd3q66uN1gS2QYXbKEIYHwySkZAIP%2BEMxXGQ2j17bxCh6VMs6BzNX60cd%2Fz70%2BQL%2FI9RBQmf4MSQgWZj5SmNw%3D%3D

**\_wb\_** _“Organic structures”_ 2022 image/jxl 44 bytes https://jxl-art.surma.technology/?zcode=TY1BDoIwEEX3c4p%2FgSZttSZsTNAFrDQRDWsCBZpUaLAaj2%2BFStz9eX%2FmTWka3yPZccq16XoPteV0ML7RLnDBKZvGpyt603ps6HK8Qig6T0YPvvJmHKDo5h7V3VkNSWRa1NiDExDiO0aAoQQTaxRLfwq94HITeTZVzVcMpoScWdxhUql5nkmZ5at3OUxfXWotZPLHCm11HVTR%2FvtLHw%3D%3D

**KammutierSpule** _“A dramatic appearance of a cube”_ 2022 image/jxl 71 bytes https://jxl-art.surma.technology/?zcode=VY7dSgMxEIWvm6cYeuFFobvtSru2UGWlImr9gQr1TtJ02gSySchOXLdv7FuYXS3o3J2P8x0mHbBXRRrnvQJ2npeclADuHHLPjUCwe%2BAgwhZZEUhaP%2B89cq8s9B94WQZS6NcuaOzDKhxRecuWnOLYsyC7RQ%2FZKMvYWh0jysewbQgrxgoDaMg3sLce7syXF4rDEkvruKemc%2BD%2B5eYW3lZQeAJhS4ekSFmTMEnk5mnalhJlsHUTRx2uIq%2FrOnE2ICUGKe0G3z1WQVOVOOmuaqmEXGT55KyWaBbdf4M0%2FqSd5Oxa0Q4dScjZk6UVr4gxtQcBl5AxgCFsUB0k4Q7Go9N1fI0Eo7b6GavTDsbQxJBP2tBe29IoCC5%2BtX9wehqL3iZ6P4U%2FTjY7WUMoPg6F1pDPziP7Bg%3D%3D

Hi all, I'm sharing the source code of my entry!

I found this pattern by exploring "large numbers" precision

I like it, looks like the output of a path tracer

yeah, I started with that idea when I found a noise pattern generator. So this image has two layers. However, I am not sure really how this is all working, because alpha channel etc, but that is the result 🙂

It is likely not a conforming bitstream given that it probably overflows int32 (and also that it's probably signaling to be ok with int16 which is certainly not true), so different decoders could decode this differently or refuse to decode it. Very cool entry though.

I should make my decoder always do overflow checking, and upon detection ~~output goatse~~ run a script to open never gonna give you up in the user's web browser

😂

or maybe this version: https://youtu.be/rTga41r3a4s

`Weighted 1000000000` it "probably" overflows i32? xD

I see, so, looking forward to see how it will look in the future on different implementations 😛

Next year I hope to participate again, hopping that the webpage has implemented the missing? "commands" from jpeg-xl

**\_wb\_** _“Blobbiness”_ 2022 image/jxl 40 bytes https://jxl-art.surma.technology/?zcode=PY7LDoIwEEX38xX3B5pQHuLKBF3ASo2PsCZ2hCYVm9rw%2FbYE2M2ZyblzW638gDRPqGHdDx4yK%2BiovWIb9ju6nR6QBV2c5tF3Xn9HlPS0v%2B5jDWNPpN8444A8JSDMV8dTwCQQIHBnwy8PIRdu5yesIFYhyiJL53vAtm42PQrV1FfGxArrZossl8zadSq2C23%2F

**\_wb\_** _“Sponge”_ 2022 image/jxl 45 bytes https://jxl-art.surma.technology/?zcode=VU9BDoIwELzvK%2BYDJLQRlYsJeoATGtFwJnSFJogEGxJ%2Fb4tA8DY7OzM7m2tlaoRbnxLWVW0QbHw6aqO4s%2FyerqcbBJ17za0pjH612NG9exfPrmFIIv1AiQN8Aiz8THAcLj0Pywx4SP%2BwoB%2FK2MCTmykhtQ4ZBNMuHzuxsgIp51wn8YQ%2FZ1kmj5PVIWeMhipqGshwxWXccOluiaVF3BfKfWbLfAE%3D

**\_wb\_** _“Autumn Macroblocks”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=PY3BCsIwEETv%2BxV7L4GkVsWLYEXwFMEqOZd2bRdqjbp68OtNQvU2b2Z2x3ErPa4WGvbEXS84LzSULC354Bdw3J5Qw%2BHBNEotfBtxCWf%2FrK9%2BIMyhur%2BIPgTAF2xwjSYKmwSiws27s5ndoYo4JWoWdCIXySSayi6zvzw6FQ3USJj5F9K3HFS4NfAF

**\_wb\_** _“Another phone wallpaper in 256 bits”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=C89MKclQMDU04vJIzUzPKFEwNDAy4XLKLElJLQBJcAU5hygYcIUWFCfmFuSkKhhxBReWpqZWpXJlpikkK9gpGHIpKACZfkCmAZCpoKCr4FiW7qftF66gawwWAMqGw2Wh8uFgeRO4kJ%2BrgjkXiBEONBAA

is it bad that I can look at this and immediately tell that it's squeezed

It is a sign that you are getting familiar with the bitstream 😅

**\_wb\_** _“North+Northwest”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=Ncq7CgIxFIThfp5iegkkYSOmEdTGahEvpBb3uBtQCGvY5%2FcIpvtn%2BM6HK33ErXzu7%2FISeuxzHaTUiRsg5UEjri2OksepMnQWyE8%2BuKUDqXmaZdEVrU7SsE80vnVDSYVxtpndMvarH3T%2F4yKVPgR8AQ%3D%3D

**\_wb\_** _“North+Northeast”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=NYq7CgIxFAX78xWnl0A2bMQ0gopgFcQHqcW97gYUwhr2%2B72F6WaGuRxudAH38n18ylvosM91kFInboCUB4WwtjhJHqdK31sgv%2Fjklh1IxfMsi1qwqqRhTDSucZuSHqaz7dktY1zFo6Z%2FuEql8x4%2F

**\_wb\_** _“West+Northwest”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=Ncq7CgIxFIThfp5iegkkwYhpBLWxEvFCanGPu4EVwm7Y5%2FcIpvtn%2BK7HO33Eo8zPTxmFHodcOyl14BZIudOIG4uT5H6oDGsL5Ddf3NGB1LxMsuiKVidpeE40vnVDSYVxtpn90qfVD7r%2FcZNKHwK%2B

**\_wb\_** _“Weighted”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=NYxBCsIwFET3c4q5QCANppiNC924ErFK1tJ8bcBCqKHn70fM7r3hMbfTnS7gUb7PuXyEDsdck5Q6cQ%2FEnBRCb3GW%2FJ4q%2Fc4C%2BcWRB3YgFa%2BLrGrBqpKGl0jjGrcoamE625r4e5Ok238ZpNJ5jw0%3D

**\_wb\_** _“AvgAll”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=RcpNCsIwFEXh%2BV3F3UAgDUbMRKhOHEnxh4zFPttACkFD1983Cc7Oge92ftAFPMvvtZQsdDilOkqpMw9ATKNG2FtcJE1zpd9ZIH345pEdSM3hK6tesLqk4TXSuNYNRRWms83069Tn%2FHd3qXTeYwM%3D

**\_wb\_** _“NorthWest”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=C3IOUTCy5AotKE7MLchJVTDicsosSUktKMlQsODiCs9MATIszQy4PFIz0zNKFExNDLi4MtMUkhXsFAy5FBSAzICi1DIgz9IAyFVQ0FXwC1fQNYKxYYrCgSp0DQ2Q1RhC2cGpJQpGpqZcAA%3D%3D

**\_wb\_** _“Northeast ”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=PcqxCsIwFIXh%2FTzFeYFAGkwxi4MidCrFWjKX9moDCqGGPr93sG7fD%2F%2FtcqcLGPJnfOeX0OGcyiy5LDwCMc2KUFs0kp5LoT9YID048cQKpLJbZdMKVpM0bCON271PUQ9T2f9z1fq5l0LnPb4%3D

**\_wb\_** _“Hairy Selection”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMDTlCs9MAXKNzA24PFIz0zNKFEwsDLhCC4oTcwtyUoHKuLgy0xT8FOwUDLgUFEBMXT9XKE9BQVchODUnNblEQdcQjW9mwgXhligYmZpyAQA%3D

I like what the upsampling is doing here, producing these 'hairs':

very spiky hairs!

yes, even though it's doing an 8x upsampling, somehow the hairs end up not blurry at all but nice and sharp

**\_wb\_** _“Fragility”_ 2022 image/jxl 38 bytes https://jxl-art.surma.technology/?zcode=Xcq7CsJAEIXhfp7i9DIwm5gFG0FtUqXwwtSSjMlCAgsuPr8j29kc%2BH7OOZXJclkQhK6XO0JHmiZ3txfqLc1LwSEKPfL7ueXV0BClF0YcEYihvoCHwQM3Iq5q5V%2BqBhg3W20saP8Dh%2FphnD6z7gZFjJG%2B

**\_wb\_** _“4K Moiré Curtains”_ 2022 image/jxl 39 bytes https://jxl-art.surma.technology/?zcode=PYzLCsIwEEX38xV3L4G0NqIbQUVwFcEHWZd2bAdqDTqIn2%2B0pbtzn1vRmqO2WNJpd0HmKEidpCssHViaVrFaWDo%2BhXstVR49HF3jq7zHjlEQyQ0V1sgISPhJaBMCBuFvDmTG3Kfc5Hao%2FLTx%2B2kyVOcTn7njSmHceLN5N2HmA3Ln6As%3D

without looking I'm guessing WP and some sort of RCT

YUP

getting easier to spot WP nowadays for me

no Weighted predictor used here

oh that says W, not WP

~~no wonder it decoded so quickly~~

you're right that it looks WP-ish, but as you can see also with the simple predictors you can get stuff that looks more irregular than you'd expect

**\_wb\_** _“Selection bias (4K desktop wallpaper version)”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMDTgCnIOUTA05QrPTAHyLc0MuDxSM9MzShRMTQy4QguKE3MLclIVTLi4MtMUwhXsFAy4FBSATD9dP1coT0FBFyhjCGUFp%2BakJpco6BpzgXh%2BCkYmplwA

**\_wb\_** _“Selection bias (phone wallpaper version)”_ 2022 image/jxl 31 bytes https://jxl-art.surma.technology/?zcode=LcnBCsIwEEXR%2FfuK9wOBJia4c6EIrlKwla6LHe1AraHO%2F2OV7u7hHtUGKTbSV7ieWvqEToe%2FQ8RF9Dkakw%2B4lU%2F%2FKpMwol5UZutN3zP3gD7Y8cAK5JrZ5fMm0q3Hb9XIJHej2%2BGnzBATvg%3D%3D

**\_wb\_** _“Carrosserie Brillante”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=Rcg9C8IwEAbg%2FX7Fu0sgiY3oIlgRnKr4QeZCz%2FZAa6hHf783CI7PU4t2XHTAmi77G2KkLJ1xs%2FJ0ZOkHRao83cunfZUno6LTJDxqq%2FIeEYjkgYwt3DIQYDhPPJu9CXC4ssKF9NNu7ptFc7ChPzNiSvQF

**\_wb\_** _“Kafkaesque Blue Corridors”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=TcuxDoJAEEXRfr7i9WaTBdloZaI0VmgAszWRESZB3OCE73fs6O4p7kW056QjjlSXLbJAUXrjIfd0ZRlGReEDPdK3e6eJUdBtEZ61U%2FnMyIjkhYgTnJ2A4b7wavYmwKFh3fTET4Xb01%2Fndah2VUQeAv0A

**\_wb\_** _“Shimmer”_ 2022 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOIKcg5RMOEKz0wB8cwNuDxSM9MzShRMLAy4QguKE3MLclIVLLj8izJT80oSSzLz8xQMuYILS1NTq1K5uDLTFMIV7BR0jbgUFHSBTCgjOLVEwZgLAA%3D%3D

**\_wb\_** _“24-bite Chocolate Bar”_ 2022 image/jxl 24 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMDTgCnIOAVHhmSlAvpkJl0dqZnpGiYKRCVdoQXFibkFOqoIFl39RZmpeSWJJZn6egiFXcGFpampVKheXrkJ4uIKhGRcA

**\_wb\_** _“Plasma Generator”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=RcyxCsJAEIThfp5iXuAg53lglUIbOyEoqUNukyxoPMy%2BPy5p0n3%2FFHNVK1JtYUJ3ezJm9Fo8Y8q4i86L8XRu8Krb8Klv4QWPn8pqg%2Bl3ZQR0Ys%2BWIYN0j%2B7GSQbfQzoYsWP%2FlMKMPw%3D%3D

**\_wb\_** _“Elbows and hands”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=Hco7CoAwEAXAfk%2FxLpAiAT%2BVoDZ2gh%2BsRVddiBp08fyK5cAUojMH3eCoKTtYGmT%2BZF1KFcu6KVwUUx%2FucQ%2BekVJ9CR86qpzHt0kWDMhgIgIMWvY8KYz9lT9r7j0SegE%3D

**\_wb\_** _“Color Leakage in the Synthetic Beehive”_ 2022 image/jxl 31 bytes https://jxl-art.surma.technology/?zcode=Zcm7CsJAEIXhfp7i9LKwuYmVENNYRfHC1sGMycCaLHHI8ztiaXe%2B8x9Ee046YkeX5gZPQXpT5vOSjizDqL99T%2B%2FulSIjo9MiPGmnMk8mkifCpnVtwB5uWxLgcOXID4UrTNbPC69Wvelb63WoY0Re%2FR0VfQA%3D

**\_wb\_** _“Untitled”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=LYzLCsIwEEX38xV3L4E%2BxZUQRXBVxQdZVzttA7GGdOj3O1B399wD5%2BCl4ygjdnQ7PpCR851SnhUVndkPo6z7Gef2EwMjp0vyPEkr%2Fjspke9xTbzY13xKCXuUBQEGdxatAardpjGNU2W2lT6rDfwWmPLPdhlsCCjqmn4%3D

**\_wb\_** _“Color Composition”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=LYzLCsIwEEX38xV3XwIxttqVUKXoKooPso40tgNaQzr0%2B03B5T3ncvYsXYgyoKbr4Q5Njru8zFbTKXA%2FCMpa0yNO%2FhPfIb%2FOicMoXvg7YkXEL1xSmJvn1KaEHdaGAIVbkNwCsnaFVdZlpTZlJos9Jt8tFSjzJ83c28K2MFVFPw%3D%3D

Random idea for a funky jxl art: encode DC with modular art once as a DC frame, then use it in many frames of an animation, each frame doing all-8x8 with AC coeffs that are the same constants in every AC ctx (i.e. they compress to a small constant size per group)

It looks like I opened demoscene channel

Which JXL-art can be reproduced in old JPEG?

with the ~same amount of bytes? pretty much none

Well, you can add arbitrary bytes to the JXL version.

JXL art relies on the fact that entropy coding in JXL can give you asymptotically close to zero size

it's essentially just a big header in front of a zero-size or almost-zero-size entropy stream

legacy JPEG doesn't have that kind of capability in its entropy encoding, so it's generally not possible

Wow. How does it works _that_ good?

ANS is pretty nice

it's based on symbol frequencies. if zeroes occur 99% of the time then only nonzeroes will end up consuming bits from the bitstream

it's a bit more complicated than that

This isn't really ANS vs Huffman, in fact jxlart will use Huffman since ANS has slightly higher constant signaling overhead 🙂

It's the MA context modeling that allows this

In jxl, if a histogram in a particular context assigns 100% to one symbol and 0% to all the others, then the encoding of that symbol is literally 0 bits. The decoder just knows every symbol in that context must have the value which has 100% chance, so no need to read any bit at all from the bitstream.

The MA context modeling additionally allows you to pick a predictor (and also an offset and multiplier) on a per-sample basis. So for each to-be decoded sample, it computes its context by traversing the MA tree (picking branches by looking at neighboring pixels), and then that context determines what predictor to use, what offset and multiplier to apply, and what histogram to use for the entropy coding of the residual value after prediction.

Now in jxl art (i.e. when using jxl_from_tree), we simply set all the residuals to 0, so all histograms have the form "0 with 100% chance, any other symbol with 0% chance", so the only thing the bitstream contains is a description of the MA tree itself.

We could by the way make jxl art a lot more expressive by also allowing nontrivial-but-very-compressible residuals. For example, anything super repetitive (say residuals that are 1,42,-82,7,1,42,-82,7,... in an endless loop) can be encoded very efficiently using the lz77 feature of jxl entropy coding. Or anything that has 99.9% chance for 0 and then some manually modified values here and there will still compress very well with ANS (here it makes a big difference with Huffman, since Huffman needs one bit per zero symbol while ANS can use fractions of a bit).

**\_wb\_** _“Lies, damned lies, and statistics”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=LYzLCsIwEEX38xV3XwJJ8bkRqhRdRfFB1pHGdkBrSId%2Bvwm4PPdczp6lC1EGbOh6uEOT4y7TcqHpFLgfBNuVpkec%2FCe%2BA2o6Jw6jeOHvCEPEL1xSmJvn1KaEXT4ACrcgOQVk6yqrrMumcHHH5LuSgDL%2FpZl7W9kWpl7TDw%3D%3D

At least I can express my current feelings artistically 🙂

lies, damned lies, and "AVIF is faster than JXL"

**\_wb\_** _“Transparency”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOQKz0wB8gwNjEy4PFIz0zNKIOzQguLE3IKcVAUjLv%2BizNS8ksSSzPw8oHrHnIKMRC6uzDSFcG0%2FXb9wBTsFYy4FBV2F4NSc1OQSBV1TMM%2BxLN1P289VwdDInAsA

**\_wb\_** _“Hardware”_ 2022 image/jxl 49 bytes https://jxl-art.surma.technology/?zcode=XU7NCoJAEL7PU8xdBHVV7BJYCJ226Ic9W226YLbo4vM3a5pth4H5fmc2ytylNjVmcNyeMQSh7oRWaQA7qaraYBIHcNF9%2BdSNxBj2nZKtKY16teTOG12XAOqBN1xjAIi0Hjo55Ne%2B6DriGHGIPp6kwShJRkQeQVIUjMiq%2BVAJj9MShl%2FOJqZG7nNOAT904ixzvBGb4KxP8EP4QiwNP2e5xwv00OUFDYMF23o237afpOBW2MD%2F528%3D

left in the ball reflower a bit too long

Just for fun, it would be interesting to see how large a lossy avif version of something like this would be: https://jxl-art.surma.technology/?zcode=ZcyxDoJADAbgvU_RnZBwqDknEzRGp9OIhvkClWuCeJ4ng09vgdGtX_-_3XJsyEeHa7jsrphBxY1IZfkSjsSti_N882_78B1hDqfA1Ecb-dmjgvL1IfoSFJ13FoDveA404EZeIaZ4CLYZ65gqsaRVYlJTSb4QT5taoCaMByV1VEtd67_VCmYXQ2sSs0eVa_gB

**\_wb\_** _“RGB splashes”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=LYvBCsIwEAXv%2BxXvXgJJNdiToCJ6iiJKzoHEdkFriKHf7wY8zrx5e64x5TphoNvhDk2eo1C%2F0XROPE4VdrD0yN%2Fwzq%2BENV0Kp7mGyp8ZhoifuJa0HEvBVt6AwqmE2BooIyyB75xyXnZldEtatFtG14lUq79w6K2lHw%3D%3D

**\_wb\_** _“Sierpinski Foam (8K, with alpha)”_ 2022 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=TY4xC8IwFIT39ytuL4FqW9FFUAedIlglczDPJlBrrM%2FFX28iIm733QfHrYM4juJR0WFzREkmuESLWUk7Dp0XNHVJp%2Fiw19gz5rQfAw9iJdwGTKi9P5lfTKs%2BeksULjhjiYaAFE2hlTaJq8SAgoaqP%2FFffqvst6N1eR3TX2VQZK%2FQsqR3bw%3D%3D

I can't upload the png to discord since it's too large

does this work?

unfortunately browsers blend images with alpha against a bright background even when you're in dark mode

data:image/jxl;base64,/wr6hsXCYAFIiB8QAEgAS6EicvNaZK0PXLNaPWQgSTgB

it looks nicer against a dark background

JXL: 33B (not MB) PNG: 35MB AVIF: 44MB WebP: 26MB

Hope you don't mind I nabbed that to test the polyfill

it's an 8K image so a bit large for a web image 🙂 but good to stress-test memory consumption I guess

that's probably for consistency across users

But yeah it would be cool if it was a flag

it doesn't make sense to me that they put `background-color: hsl(0, 0%, 90%);` on images you open in a tab, making it have a different background color than the rest of the tab

I made a smaller web-size version <:PepeOK:805388754545934396>

The smaller version is 35B while the 8k version is 33B <:kekw:808717074305122316>

that gives me a thought, jxl images could eventually have a flag or something that "recomends" dark or light, smooth or checkerboard alpha backgrounds for the image.

actually I think it could be a good idea for image formats in general, not just jxl

then again, potential feature creep?

PNG has a chunk for that but I think nearly everything ignores it

preferred palettes?

That too but you can also select a background color to blend to

ah bKGD

Anyway in jxl you can also just add a background layer with a solid color, which will force viewers to use that (they will not even know, libjxl will do the blending for them by default)

(while you still preserve the original rgba image in the foreground layer)

PNG is better than AVIF

**\_wb\_** _“Having fun with the nonseparable upsampler, again.”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=LctBC4JAEIbh%2B%2FyK7y4LrrbgKagIOm1Qyp4lJx0oXWrw9zdCt%2Fc5vEfRgbNO8HQ7tfCBkgwb60AXlnFSVLuSuvzt3%2FnFaOj6EZ61V1lmm0ieeGCPkgDLVEQXkzmYAQeD8%2F8%2BrGMs4hkVbbqz2vUD

**\_wb\_** _“Superhero Comic Book”_ 2022 image/jxl 31 bytes https://jxl-art.surma.technology/?zcode=NYq9CsJAEAb7fYqvl4WsZ8DKQglYRfGHqwNZzYLG41x8ftfCbmaYrfmoxSckOu0ukJayjaGSWtqr3SfHctXQtbyHZ3ko1nSoprMPbq8ZQmQ3HKt%2BulqxAQsBUfKi5z5HaMIBRginP3eQ38c4q8fxBQ%3D%3D

**\_wb\_** _“Ceci n'est pas 3D”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=TYoxC8IwFAb39yu%2BXQJNxWAWB6XgFEFbMpf2aR%2FUGvQN%2FnwDttDtjrujaM9JB9iCrqcadkdR%2BuzeezqzPAaFc46a9GmfaWSUdHkLT9qqvCZYIrmjwwEFARm%2FMwIGMeeFTPnvcRNMiKvpxiN3CrOdPVTY0w8%3D

I just love how a lucky tree using just simple predictors can create things like that

``` if W+N-NW > 0 - Select -3 - NE 8 ```

Change those numbers even just a bit and it does something quite different and not really 3D like, but for -3 and 8 it does produce something nice

**\_wb\_** _“Variant”_ 2022 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=NYrNCsIwEAbv%2BxTfXQJNi9FcPChCTxH8IefSrnah1lD34OMbMN5mmNmLDpx0hK3ofLjCrinKkN17Ty3LY1Q45%2BiW3t0zTYyaTovwrJ3Ka4Ylkjt67FARkPFTEDCIOf%2FJ1L8eV8GEmCdjm02pF564V5imeDhiS18%3D

Changing the number in the condition makes variants, negative numbers give nice deep shadows

what does the piece of code represent ?

which language is this ?

you can redirect me to any other piece of information if you don't want to explain again 😄 I feel like I'm coming late on the subject

it's a representation of the meta-adaptive context modeling trees we have in modular jxl

https://jxl-art.surma.technology/wtf.html this gives some info on it

thanks 🙂

**\_wb\_** _“The Anonymity of a Depressing Big City”_ 2022 image/jxl 38 bytes https://jxl-art.surma.technology/?zcode=ZY3BCsIwEETv%2BxVzL4GkNdiToCJ6iiDKnouNbUBrqEH8fDeivXh7%2B2aHWYXU%2Bph6GE2H9RHGEodW7lmtaedD1yeUc02n%2BGhu8epR034MfkhNCvcBhihccMZCCBB8CWpBQIFRTqS%2BORdOOZYnZT9ZVopZXOGYp3IubcemzUtQVfkvq9%2FK8tm5wm1gjaE3

**\_wb\_** _“Folding an old-fashioned map”_ 2022 image/jxl 43 bytes https://jxl-art.surma.technology/?zcode=XY1NC8IwDIbv%2BRXvfRTaqXgTVARPFUTIWVzdctlEiui%2FN26lWnN68n4kG4lNuMUOztLhLqGP5yhDD0cszSjXc9oHabs48XF7UpPkigtWsAQkdIrj8kz6Zww4GRObnHoVKV%2BwoxxZuNxgw2w8V55ZDfM9u360XlVUmP1pO6Vl7lde68VbTan6e4tRW0tv

i'm not aware enough of jxl/programming but have been playing with the art and wondering about differing implementations ``` if N-NE > 0 if c > 0 - Weighted - 44 - Weighted + 6 - N - 9 ``` produces jxl 28 bytes but a different png when downloading/viewing from the site VS viewing jxl file in browser/ImageMagick

That image will go into very much out of range values and overflow even int32_t (while the values are assumed to be in uint8 range). It's technically not a valid bitstream.

It's somewhat strange though that different libjxl integrations show different results though.

**oxyzoron** _“Color Bars”_ 2022 image/jxl 70 bytes https://jxl-art.surma.technology/?zcode=rZA7EoAgDER7TrEXcEbioENjbW%2FhAfzSU%2BjtBUVJxtZuH5sNC4Ob%2FAYytepmt24e2pJSbsGBFtpYBQTYA5RBFhhQ6Ev0sw8xE0fH5CYZ%2FTdVCaorTtZw0tQIvO7OaMUionvT3aNkOnb62xEN85D4hPQ8NsqaP5mc%2BJ6c

it isn't there anymore?

forgot to add it back, it should be there now

still getting 404, guess it takes a while?

https://grass.moe/catmull_rom/ ah wait it's underscore

(for a recent `jxl_from_tree`)

https://embed.moe/https://cdn.discordapp.com/attachments/824000991891554375/1056150955449663498/jxlmas.jxl

data:image/jxl;base64,/wq6IUWRGGggAAC4AGY008xMmjVjBgBLOELmtusLakzkRUU6YGGP940I+Buwq1g21A3q142BCSQ4wArYY1QAAIADAABAAARAAAQARgLASSJlIwxR4sJ4mXe8mMcQPgCAq7BilFisFfBjAAsIpWlJDEHrzGxO7AnH/Cjh0Fdl3zaKadCciBO7Js0pICb/13YEFCUHBXDyaQZIWhvPt8MDK71WzJkut6XdnEsFeQhLGwQH7TvBAeBfysWa+xuH1oeF+hPinAXWr7h3hwNIYLyaIsYUj2Exd+GTWaxOZ9mcKMZ+vMxLSHqmeTjooa3xsWeDlWwUPX3CNXYUHczidTiM4dprg0Z2DdIDYWU62aMKY5kHMCZI+tmj+FJpXm4+iULslS+pAMQYL4HgACsDAwMD

Thanks, that was useful to draw my merry jxl-mas card 🙂

a more saturated version and the 'Merry' a little less fat

data:image/jxl;base64,/wq6IUWRGGggAAC0AGY008xMmjVjBgBLOEPmtusLqDGRFxXpgAUT8H4jCvwbsKtYNtQN6tONBAdYAdhjVAAAgAMAAEAABEAABABGAsBJImUjDFHiwniZl7yYxxAuAICrsGKUWKwV8KMACwilYVEMQWfMbE7sSXzKKOHQV+U5KIpp0JyIPokmzSkgRmY6dwQUJS+5AfNpBkhaYqq3wwMrPaKkmS63pd2cS8FXCEsbBAdhosMBdFzK6yzwDTrrw0L9CXHOMu9V3LvDASTgX00RY4rHsJh7sf4sVqezbM60XBRe5iUkPdM8HPTQ1mTas8FKNoq1H+EaO4oOylM7HMZw7bUR2EUD6YGwRiLsUYVxTAMYExEH7FkUqDQwBLNEIfYuW1QAkoyXQHCAFQMDAwM=

https://embed.moe/https://cdn.discordapp.com/attachments/824000991891554375/1056249789592907827/jxlmas2.jxl

**\_wb\_** _“Snow Crystals”_ 2022 image/jxl 27 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKz0wB0pZmBlweqZnpGSUKpiYGXKEFxYm5BTmpCiZcwYWlqalVqVxBziFA1VyZaQrJCnYKBlwKCkBmuLafrl84kK9rDBRQUNBVCE7NSU0uUdA1hPNLFIy4YCwDLgA%3D

**\_wb\_** _“Curvature”_ 2022 image/jxl 36 bytes https://jxl-art.surma.technology/?zcode=Nc2xCsJAEATQfr9i%2BrCQu8RWUJGk2kKUq0PuTBZUjniIn58tku4xMzBnLTHlMsPVFDQaDs5Tn3Sai2W%2BpUf%2BDu%2F8SvB0u9zhiEifGHE0Asa%2FsTYCjAD2G2XvQyUswUbc7F23DFHTp4DbLTn9Jqnkau%2BOVg%3D%3D

**\_wb\_** _“Untitled”_ 2022 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKLShOzC3ISVWw4ArPTAEKGRpZcHmkZqZnlCgYmZpxBReWpqZWpSpwBTmHAJVnpilUKNgpGHApKACZflCmgoIukK1rCGYCxcOB4rrmYB5IKhwmBeIEp5YoGJpywZgGXAA%3D

**\_wb\_** _“Batteries?”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKLShOzC3ISVWw4ArPTAEKGZmacXmkZqZnlCgYGllwBTmHKJhwcWWmKfgp2CkYcCko6Co4lqX7afuFK%2BgaArlAmXCgjK45kA2VDNf2g8iB%2BMGpQINMuQA%3D

**\_wb\_** _“8x something”_ 2022 image/jxl 26 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKLShOzC3ISQUywzNTgEKGRhZcHqmZ6RklYGaQc4iCARdXZppCsoIdkAVkhIMZugrBqSVgOjxcwdAULKEbDpFTUIDJglhAQXMuAA%3D%3D

**\_wb\_** _“Simple rule, unexpected outcome”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOEKLShOzC3ISQUyg5xDFAy4wjNTgBKmhkZcHqmZ6RklYCZXZppCuLafrl%2B4gp2CrrERl4ICsoixIVBAQUFXITi1REHX0AzOy0lNLlEASUKkgDIA

This one was a nice surprise

**\_wb\_** _“Simple rule, unexpected outcome (2)”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOEKLShOzC3ISVWw4ApyDlEw4ArPTAFKGJmacXmkZqZnlICZXJlpCuHafrp%2B4Qp2CrrGRlwKCsgixoZAAQUFXYXg1BIFXUMzOC8nNblEwZgLJgWUAQA%3D

**\_wb\_** _“Simple rule, unexpected outcome (3)”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOEKLShOzC3ISVWw4ApyDlEw4ArPTAFKGJmacXmkZqZnlICZXJlpCuHafrp%2B4Qp2CrrGRlwKCsgixoZAAQUFXYXg1BIFXUMzKM%2BxLB2oRAGkGiIFlAEA

**\_wb\_** _“Simple rule, unexpected outcome (4)”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOEKLShOzC3ISVWw4ApyDlEw4ArPTAFKGJmacXmkZqZnlICZXFyZaQrh2n66fuEKdgq6xkZcCgrIIsaGQAEFBV2F4NQSBV1DMyjPvSgxJTM1r0QBJA2RNDThAgA%3D

**\_wb\_** _“Simple rule, unexpected outcome (5)”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOIKLShOzC3ISVWw4ApyDlEw4ArPTAFKGBsYcHmkZqZnlICZXJlpCuHafrp%2B4Qp2CrrGRlwKCsgixoZAAQUFXYXg1BIFXUMzKM%2BxLB2oBCjABZMzNOECAA%3D%3D

**\_wb\_** _“Simple rule, unexpected outcome (6)”_ 2022 image/jxl 31 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOIKLShOzC3ISVWw4ApyDlEw4ArPTAFKGBpZcHmkZqZnlICZXJlpCuHafrp%2B4Qp2CrrGRlwKCsgiZsZAAQUFXYXg1BIFXUMzKM%2B9KDElMzUPKGTKBZM1suQCAA%3D%3D

**\_wb\_** _“The Big Puzzle”_ 2022 image/jxl 38 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKLShOzC3ISQUyg5xDFAxNubjCM1OAUkamZlweqZnpGSVgJldmmkKygp2CIYhRAWQYcOkqhAO5IFIXLBqu7afrFw6U0jU0suBSUEAWMjI1BYooKOgqBKeWKOiamUB57kWJKZmpeUAhkA6ILEgtAA%3D%3D

**\_wb\_** _“Festivities”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKLShOzC3ISVWw4ApyDlEw4ArPTAFKGJmacXmkZqZnlICZXJlpCuHafrp%2B4Qp2CrrGRlwKCsgixsZAAQUFXYXg1BIFXUMYz70oMSUzNQ8kBNEQUJRa5lpUBNRhgKTB0BSZY8IFAA%3D%3D

**\_wb\_** _“Diagram”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOIKLShOzC3ISVUw4QpyDlEw4ArPTAFJmJpxeaRmpmeUgJlcmWkK4dp%2Bun7hCnYKukYmXAoKyCLGhkABBQVdheDUEgVdQzMwD1mBAVgEpMK9KDElMzUPqMwIi5gpF8wUoJEA

**\_wb\_** _“Tinfoil halftoning”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=Tcq9CgIxEATgfp9iellI7lSsLBQ5qxT%2BsPVhVm%2FhPILm%2FTEGketmvpmd5agpD%2FCOrundP9Oo2NBpf4EjsViWlW%2FoqPYYco1kd8gicBBswa0jYC6tLwAwRMB%2BXct8d1W%2Bh%2B7VR9Mpl9vfzjrq7SeMcECzpA8%3D

**\_wb\_** _“The decay of gold”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=Tcy9CsJAEATgfp9iejnwNhymEtTCVCn84ergrcmCyqGLz%2B8qFumGb4bZqhWpNqGlc30N93oTj4fdCZEoa%2FEmRaZOdJzsF0mvuGDtfcBRDMsv5EUf%2BuwaIrcEzIlTcgECNu%2FRFYGZ%2F7J%2FDkXlYf42G3Czog8%3D

**\_wb\_** _“Mondrian Revisited”_ 2022 image/jxl 37 bytes https://jxl-art.surma.technology/?zcode=TYs9C8IwFEX39yvuLoEkNFAXQR3aqYgfZC7m2T6oEmro7zcpDt3uOYd7khQ4phE1PeK3f8eJUdH1fIcm8hJyccZSyzKMaZ0kL1xmXnCA2utCftepzhc2tiZgq6xz2QAKx2XIFsoa%2FTc3nviZ8osKNXMfhD9pvWzjDw%3D%3D

yo that looks great

gold bismuth vibes

**\_wb\_** _“Bruges Lace”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOQKLShOzC3ISVWw4ApyDlEw4ArPTAFJGFlweaRmpmeUgJlcmWkK4dp%2Bun7hCnYKusZGXAoKyCLGhkABBQVdheDUEgVdQzMoz70oMSUzNQ8kZIqkAMgBAA%3D%3D

**\_wb\_** _“Untitled ”_ 2022 image/jxl 31 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKLShOzC3ISVWw4ApyDlEw4ArPTAFKGBpZcHmkZqZnlICZXJlpCuHafrp%2B4Qp2CrqGZlwKCsgixoZAAQUFXQU%2FVwVdIxMoJzg1JzW5BKgcJhmuAJQDAA%3D%3D

**\_wb\_** _“Untitled ”_ 2022 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKLShOzC3ISVUw4gpyDlEw4ArPTAFKmBoacXmkZqZnlICZXJlpCuHafrp%2B4Qp2CrqGZlwKCsgixoZAAQUFXQXHsnSgoIKusRFUIDg1JzW5RAFN3tCcCwA%3D

**\_wb\_** _“The tablecloth with the weird pattern on the diagonal”_ 2022 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=LYqxCsIwFEX3%2BxV3l0CTosnkoItTEbVkLu3TBloIGvr9vpZu53DOJZVBchnp0eZfN%2BdJGPC4vlgBMQ1ajtbhJukzlg2B9GbPsw4K968syj7A8CmF1vmdjHVhPeKhMU3U51SDXNskvea62lSTwx8%3D

**\_wb\_** _“Untitled”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=TYm9DkAwGEX37ynuLk1aIUwGFpPBTzoLH5qQVNTi6VUkYrrn3JMbN7J1CyLq7NFvdmWEVBctJGkz%2BqBkGFHJZl7cy81%2BMl8MIjNhQAb1gA4qUWlvIiXgfyTeAYGGHUT8ycqDd%2FWLMb0rbw%3D%3D

**\_wb\_** _“Chessboard Madness”_ 2022 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKLShOzC3ISVUw4QpyDlEw4ArPTAFKmBsYcHmkZqZnlICZXJlpCuHafrp%2B4Qp2CrqGRhZcCgrIQkampkARBQVdheDUnNTkEgVdI1MzVBETLiSOkYkBFwA%3D

Onion Cells!

**\_wb\_** _“Untitled”_ 2022 image/jxl 36 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKLShOzC3ISVUw4QpyDlEwNOXiykxTCChKLXMtKlKwUzDgUlDQVQhOzUlNLgFzgLLh2n66fuFASV0jUzOgEKqgkakpWAxJH1wdkpgRFyrXyJgLAA%3D%3D

**\_wb\_** _“Untitled”_ 2022 image/jxl 37 bytes https://jxl-art.surma.technology/?zcode=TczLCoMwEIXh%2FTzF2UvARKSuhNZ9kF7IWnRaBYXQBp%2B%2Fx4LU3c83l8uUBo1pRCWP%2BOmWOCucXJs7rMj0RI%2BaxWjfurJzMTivr5B5GMu%2BaaJxTjE%2BcIMMHMG6EwXYDwNM8Qf%2Fg%2B1mezZrn%2BDKUr4%3D

**\_wb\_** _“Happy New Year (no, that title is not related to the actual thing)”_ 2022 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKLShOzC3ISVUw5ApyDlEw4OLKTFMIKEotU7ADchQUdBWCU0vALKB4uLafrl84UMbI1BQoApINV9A1MjWDchzL0v20%2FVwVjLkA

**jjido** _“Reticula (idea by _wb_”_ 2023 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMDLk8i%2FKTM0rSSzJzM9TMOHiykxTCNf20%2FULV7BT0DUx5VJQQBYxMjUAiigo6CoEp%2BakJpco6BqZGqOKGHEhc0wMuAA%3D

**\_wb\_** _“Dunes”_ 2023 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOAKLShOzC3ISQUK%2BBdlpuaVJJZk5ucpGHKFZ6YAlRmZGyh4pGamZ5QomFgYcHFlpimEa%2Fvp%2BoUr2AE1KyjoKrgXJaaA9CnoGhqZoYoYmZpyAQA%3D

**\_wb\_** _“Lines”_ 2023 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=NYq7CoAgFIb38xT%2FHgdSqDGodoMonAWtDnSRkp4%2Fl8bv0knyIaYNisZ%2BQklzfNwR9wBNwy3hTC7JdeZsxeetUppIFtjCsLFowKomgNG%2BqymyYf1jXqCqkj4%3D

**DZgas** _“tape”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOAKLShOzC3ISVUw5PIvykzNK0ksyczPA%2FLCM1OAygwNjEwUPFIz0zNKwGwursw0hXBtP12%2FcAU7BWNjQyNjLgUFXQX3osQUkG4FXUMwQBU0MjDgAgA%3D

**\_wb\_** _“Shades of deep purple”_ 2023 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOYKLShOzC3ISVUw4fIvykzNK0ksyczPUzDkCs9MASozNTHg8kjNTM8oUbA0M%2BAKLixNTa1K5eLKTFNIVrADKgMywrX9dP3CgTxdYyMuBQVdBceydKCYgq4FmBeuYGbMBaLMAQ%3D%3D

**\_wb\_** _“$$$$$$$$$$”_ 2023 image/jxl 29 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOQKcg5RMOYKLShOzC3ISQUK%2BBdlpuaVJJZk5ucBeeGZKSBlBkYmXB6pmekZJRB2cGFpampVKhdXZppCpYId0AQFBSjTAshUUNBVCE4tUTBAYusaG3IhxAE%3D

**\_wb\_** _“Lines”_ 2023 image/jxl 25 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOYKLShOzC3ISVUw4vIvykzNK0ksyczPUzDkCs9MASozNDAy4fJIzUzPKIGwgwtLU1OrUhW4uDLTFPwU7BQMuBQUdBX8XBV0DWEsY0MuAA%3D%3D

**\_wb\_** _“Interference”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOQKcg5RMDTlCi0oTswtyElVsODyL8pMzStJLMnMz1Mw5ArPTAGqMzI14%2FJIzUzPKAEzgwtLU1OrUrm4MtMUwrX9dP3CFewUDLgUFHQVgExdQwjLVcHYkIsLAA%3D%3D

**\_wb\_** _“Moiré spotlights”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOYKLShOzC3ISVUw5PIvykzNK0ksyczPA%2FLCM1OAykwNjbg8UjPTM0oUDA2MTLiCC0tTU6tSubgy0xTCtf10%2FcIV7BR0DY0suBQUdBUcy9L9tIFCuoYwbjiIa2RqygUA

**\_wb\_** _“Feathers”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOYKLShOzC3ISVUw5PIvykzNK0ksyczPA%2FLCM1OAygwNjEy4PFIz0zNKIOzgwtLU1KpUBS6uzDSFcG0%2FXb9wBTsFXUMzLgUFXQXHsnQ%2FbaCIriGYC2QZGwIA

**\_wb\_** _“Skyscrapers”_ 2023 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=XYvBCoJQEEX38xX3BwTn%2BaBdi9q0C7JoLTjlQNmzJki%2FvqFA1N053HM3arUkaxDpsD2ioFN6Vfd0EzDtnyqtVaaP1u2stWech0g70Wtjfy67t8ggRHpBjzUCAY4fR47OoxX8MyBDKX6eWcbTdLXY4sxzGpdAXw%3D%3D

**\_wb\_** _“Iron Curtains”_ 2023 image/jxl 40 bytes https://jxl-art.surma.technology/?zcode=ZY3LCsJADEX3%2BYrsS2CmRtGNoCK4GsEHsy52bANaBw1%2Bv6HVbtzdnHMvWYvWKWuL3sFhc0IP5%2Fyq7vmWkGH%2FlNRppfLoTESprTdlB7skTau4mDkAueIFl6YRCY9JkSal5S92Fkfhy3l%2FmotFoBCtQAPqoaHtuBl2q3cTCqPEfzDaJ%2FghqzAzfAA%3D

**\_wb\_** _“The Matrix”_ 2023 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=Tcq7CsJAEEbhfp7i78NC4sagjWBEsFohKlsHMiYDGlcdU%2BTps14Ku8PHKUUbDtphQdXmiJxO4Vlfw4Vj7h%2FCvdYqtx4ZeWnitixS2rG0nWKep3S4v5hHJpIzfOKM81jB2BkB%2F2KzCIDBemhdEum7%2FMR%2F5L0YuC0KSxM%3D

**\_wb\_** _“The Matrix (phone background)”_ 2023 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=Tcq7CsJAEEbhfp7i78OCuxvFSlARUq3gha0DGZOBGDc6Wvj0rpfC7vBxVqINJ%2B0wp936gJKO6VafU885t1fhQWuVywBLUZq8Ta2jiqXtFHbiStqPd%2BYnE8kJsQgmRCxgvCPgX7zNABgsH20oMn2Xn8SPvBeDsMHM0ws%3D

**\_wb\_** _“Ripples”_ 2023 image/jxl 34 bytes https://jxl-art.surma.technology/?zcode=TcqxCsJAEEXRfr7i9bLgOi5aCZrGKgE1bB3MmAzEuJgh3%2B8GRGzPvSe1VpL1CHQpbmCq09Q80yDYU%2FVWGa0xfY3wFLXNm98FOot2vYHDmkgfiKvSlREHON4Q8C%2FsMwAOx7nL%2BD0WuMogd4Nbhl%2FmLX0A

**\_wb\_** _“The New Variant”_ 2023 image/jxl 39 bytes https://jxl-art.surma.technology/?zcode=TY49C8IwEIb3%2FIrbJdAkLdTFQYc6VbFK5tCcNlBriIe%2F30sr0u197v3g9oE8RhqgEpfDFZS4xbd7xhGhFKcUcCJH4TWxYYPnmCp0KY4YHgMtWoQ79LDjAADLc8IPU8EEIKFDAlmvoDJLzm5a2VpOSr2d7fXNmPmUO8xS13%2FscMSeJ9VvsknO5ydz5Qs%3D

I have no clue where this one comes from, I was just trying random stuff and somehow found it. It's quite fragile (at least the part involving the first two channels), changing the constants even just a bit ruins it.

**\_wb\_** _“The New Variant (simplified)”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOEKcg5RMOAKLShOzC3ISVUw5PIvykzNK0ksyczPA%2FLCM1OAygwNjEy4PFIz0zNKIGwuBYXMNIVwbT9dv3AFOwVdI0ugCKqYsTFYSEFBVwHI1zWygHODU3NSk0sUdA25IHz3osQUkJ0gLQA%3D

**\_wb\_** _“What is this”_ 2023 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=TY0xC8IwFIT39ytul0ASLHVy0EGnClXJHMzTPmhraN%2F%2Fx0QXt%2Fu%2BO7iDaOKsA3bUH2%2BwdM9rnPLIcHRZhGeNKu%2B5UJBUZs76LZ1ZXoP%2BMskTYdOZLmAP46wl4F%2F5pikGMCho%2FLevdOWRHwrTUqXTElM9g28tfQA%3D

**\_wb\_** _“Also: what is this?”_ 2023 image/jxl 37 bytes https://jxl-art.surma.technology/?zcode=ZY09C8IwFEX39yvuLoE0baQuDjroVMEPMof2aR%2FUGur7%2F5hiB8HxnHvg7kQ7TtqjpvP%2Biopu6R2faWA4Ok3Co0aV14iCgnQ526wtHVkevcJXlkjuCKvGNAFbmMLVBPwq5302gEHGZZ%2FpwgO3ms23b3Nql%2BkwxW7%2BRWndvyrpAw%3D%3D

**\_wb\_** _“Or this?”_ 2023 image/jxl 34 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOEKLShOzC3ISVUw5OLKTFMI1%2FbT9QtXsFPQNTSy4FJQQBYyMjUFiigo6CoAuVB5EM%2B9KDElMzWvREHXiAuFb2RpzgUA

I like these names

looks more like a saw edge

some kind of beaded threads

that is _really_ cool!

**\_wb\_** _“Strange checkerboard”_ 2023 image/jxl 27 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKcg5RMOAKLShOzC3ISVUw4uLKTFMI1%2FbT9QtXsFMwMjXlUlDQVQBydC3ArODUnNTkEgUjEwsuAA%3D%3D

**\_wb\_** _“(same image as above but without the 2x upsampling)”_ 2023 image/jxl 27 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKcg5RMOAKLShOzC3ISVUw5OLKTFMI1%2FbT9QtXsFMwMjXlUlDQVQBydC3ArODUnNTkEgUjEwsuAA%3D%3D

**arborelia** _“cave crystals”_ 2023 image/jxl 46 bytes https://jxl-art.surma.technology/?zcode=XY3BCsJADETv%2BYq5y0JWVvQkqNSjLSrsuWjUPdhKuvT7jWtB8JZ5M48cd2csaZvyVV75Ac9Ua5Iutzn1HQLttX1K0w9wYcFwfsVE6YYL1vAE2NmojJWqAReM%2FLN5YYDDoTL%2Fl6Klb%2BkQP4%2BLGc3hiW7Ge5yV3QROkm3I9AY%3D

I like that style!

thanks! I am having a lot of fun with this web app

**\_wb\_** _“Untitled”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMDTiCnIOUTDgCi0oTswtyElVMOEKz0wByhiZmnF5pGamZ5SAmVyZaQoBRallrkVFCnYKxoZcCgq6CsGpOanJJQrmQA5QOlzbT9cvHCirC9KgoABSAeTrmiDYRqamXAA%3D

**\_wb\_** _“Kryptonite”_ 2023 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOQKcg5RMOQKLShOzC3ISQUyuTLTFAKKUstci4oU7IB8BQVdBcey9HBtPwVDUyAPKA1k6%2FqFA2V1zUyAIkgqdGH84NSc1OQSBSMjIy4A

**\_wb\_** _“Count the triangles in this picture”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=C3IOUTDgcsosSUktKMlQMOEKLixNTa1K5QotKE7MLchJVTDi4spMU0hWsFMw5FJQADLDtf10%2FcKBfAMgHyIC5OgagnkKCroKwaklQK4ZCh%2BiFsIGqYSJAgA%3D

**\_wb\_** _“Signals”_ 2023 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=C3IOUTA04HLKLElJLSjJUDDmCi4sTU2tSuUKLShOzC3ISVUw4uLKTFNIVrBTMORSUAAyw7X9dP3CgXwDIB8iAuToGoJ5Cgq6CsGpJQq6RihciFII25gLxAoH6QAA

**arborelia** _“Rift”_ 2023 image/jxl 37 bytes https://jxl-art.surma.technology/?zcode=c06KdC7icsosSUktKMlQMDTk4spMU0hWsFMw4FJQADIDilLLXIuKgAK6JkARdDEjsJiCgq6Cn6uCriGCFw7kmXFBOOEKhlDTwqEGg0Qdy9LDtcHqoALBqSVAhQZcAA%3D%3D

nice algae

**Kampidh** _“Pins n Bricks”_ 2023 image/jxl 39 bytes https://jxl-art.surma.technology/?zcode=Vc7BCsIwEATQ%2B37F3EsgCa14ElRETxFE6Tk0sV2oNdTg95tUCHqbGd7CtuziAN2s6OS5H%2BISL%2FsrtKTzzH6KNvJzQk07js6HhNd0Cy%2F7CKOHJuI7OmygCEixrYwwbeoydUDAQOimlAOU%2FMKumN8rtSyZbt%2B9qbIvy3G2Lv8DXdO%2FEYo%2B

HEY BRO NICE BRICKS

Brown bricks

**\_wb\_** _“The Forest”_ 2023 image/jxl 34 bytes https://jxl-art.surma.technology/?zcode=C3IOUTDhcsosSUktKMlQMOcKLixNTa1K5eIKLShOzC3ISQVKh2emgOTMLLgy0xSSFewUDLkUFIDMcG0%2FXb9wKB8s4gfkGIA5Cgq6Co5l6UAlCrrGyCJ%2B2kAtukZcqAIgvq4CiAYA

**\_wb\_** _“Advanced Trigonometry”_ 2023 image/jxl 34 bytes https://jxl-art.surma.technology/?zcode=PY2xDsIwEEP3%2Bwrv1UlNgBUJGGDKQEGZq%2BZoI7UotIGBr%2BdaEJufLdvnwwXW0D7mICl32FD1eIq8heiapnpIvcCSj2HOjKWTxLbLMKVdU7yhwRaGAJW%2BcOz8jxfHKZQLAIxKemkyePV3dq%2FWFVrhb4NxHOsQ5Z71cUavUx8%3D

**\_wb\_** _“Yet another phone background”_ 2023 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=C3IOUTDmcsosSUktKMlQsOAKLixNTa1K5eIKLShOzC3ISVUw4grPTAHKmRoacWWmKSQr2CkYcCkoAJnh2n66fuFAviGQDxbxg0qCgK6CY1k6UImCrgWaSDhMCCzgBxIw5gLxwhXMuQA%3D

**\_wb\_** _“Ripples”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=NYrNCkBAFEb39ym%2BvaaYkJVCspP8ZC0uptA0Js9vLOxO55xc2YW13ZFQW%2FQIIhr0PZ36YAREakVj%2BCmNQQqfAIHs2Uav%2FkbAZceiHl0VcejMd1RmWhRfFuI3HR88W0gp6QU%3D

(some "airplane art" above as I am traveling back from Sydney)

**\_wb\_** _“Blue, Yellow and Red”_ 2023 image/jxl 41 bytes https://jxl-art.surma.technology/?zcode=hY3LCsIwEEX38xWzL4EktDYrQV3oKogPZl3M2A5UCTX0%2Bx0Qqjt3c8%2B9hzntLug9bKUkzmXAANf86h55ZD1JkiLfWjiw9EPBOlgAueNx4hnX6HwLiAbPPPKtoFnVGrWmKppIOrCaP8QQKaoiffGvGRaymXudoftn7qcuCT%2FVdQvTn75p4A0%3D

I got my trophy from the Inercía demoparty! Nice!

same! the jxl detail is so cute

Is it possible to make animations with https://jxl-art.surma.technology/ or does it only work for static images?

good question — I think it's still using a pretty old version of libjxl, not sure if that one already has the syntax to make animation

besides having syntax, it would need to have the js decoder adapted to render animations

(I would also update it to use native jxl decoding if available)

if anyone wants to attempt to do those updates, that would be cool

Thanks for the clarification. Seems like animations should come in the future. Just when people have the time and skill to add them. I’ve still got to learn how to make static art with jxl so that’s fine with me I was just curious if it was possible.

**MSLP** _“Polish Air Force checkerboard (simplified colors)”_ 2023 image/jxl 61 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOQKz0wB0iYWXB6pmekZJSBWkHOIgiUXl76WQhFYKK0oMTdVQUufKzNNoULBTsHEmEsByKwEMo2MQaxkIMtQQVchOLUEThtwwQSQTcpMzUkpVlBTKMkvUNBWSMovKcnPxTDfCGG%2BCT7zEW6A2QQTUsCnDcVdOalpxDoLxVXoNgIdQdilCkj6cKtGOAvJAVg8ilU3AA%3D%3D

**MSLP** _“Zigzag”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOQKz0wB0UYGXB6pmekZJQrGRlyhBcWJuQU5qQoWXFyZaQrh2n66fuEKdgq6FlwKugp%2BCromIDo4tUTBBAA%3D

Just wishful thinking, but someday what if there was an image search for JXL art. Put in your original image, get one that's similar enough but only a few dozen bytes large

**MSLP** _“Mountain peaks?”_ 2023 image/jxl 25 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOEKz0wB0oZGBlweqZnpGSUKZgZcoQXFibkFOalAaa7MNIVwbT9dv3AFOwVdcy4FXQXHsnQ%2FbT9XBV1DBC9cwdCQiwsA

**Anonymous** _“Unnamed piece”_ 2023 image/jxl 76 bytes https://jxl-art.surma.technology/?zcode=C89MKclQMDEw4PJIzUzPKFEwNTPg4spMU6hUsAOzFRSAnAogB8TUVQhX0DUEM4JTSxSMTE1BSpOhslAmSB6uy9QcmWdoaILCNUdRa2RkgcK1MEXmGpsYoXAtLcFciEsMkNggV1FbBsWFCEUowQDzIJJaJKfDNCG0YIoAAA%3D%3D

**privateidentity** _“Noonle”_ 2023 image/jxl 1906 bytes https://jxl-art.surma.technology/?zcode=7VpNct0gDN73FO8CneHHYLzpJXqETjuTfRf09s2b2BJjhJCwkzwn2TF6CMSnTwLr6enP7dftx818e3oZ2Pvg3ya5D7zbRnHaRkvcRtYlGEbQsQsoOQdaLoCaSwnXx60CqPkEapMFtWkCtSmBWrCgFiZQCzOoRZPwDKAWZ1CbDajNHtTmCGrJgFpCQBIikhCSBSFZniH5fvv5%2B%2B%2FNhqUYPf%2Bc8TgZz5DxDHk1sdLfRM7ZTUSt%2FYLPq6y9uiyvzqlWibXKSo6MXs7o5bxiuV%2FI7Bc0xMIuEXusBMxIwPbONFIrZzJyJiMlMlIiIw%2FyyoPWMdwUDkkABolaAdELNfdgIUINWGyUc8mQsODG95iQwxLmbbJvznV%2BabFDD0bBkTVJnY5LSSIfKT6NosVJECURgkpiIe0LVLtB6DqZoMiKBcA0lGsOZ0LTOyJKF3Q9laKJ0FucJEBhlih4JcRl9u05gox12hEFZUUZbhRGpNzhPFdjrcCjx9H1ebJnmJmIEKbJpoaGO6HFfBArkS78D4OE6bFEpgfHIf4UQSDjD2BjmhIhRoP0Ke6QbrbD66S8Q%2BYoJqAm8R0KXucleY5grWmzVylqO5TNBxo3Yug3LiLyIm8gf%2FxO56hrNHN6T2aavwUcJZXp5xCJV%2FdNPUpw4qUNPFbe4AE4u0x93xCz3dR%2BqVZGDX69NFyBWaWRSnrP0QJTiqQEpttZE55oH6DETzVGRNqJA98vfbqSGHXp2kWOZq78ncmJqJxQ4Y5UNC1vUY5oa%2FGOoDFvpEnE5p4Qu6xqC%2BhwYcOfLjywdtDHvMeGche63kDTqrzmkWH31KehilwkvvlIP55kFxGqRXz27lyqbHXACDJ1NG6vNzJCsIpm9FXPfZV6rg%2FFaKCeW%2BjD%2B3pz22yJtRX1XO3adD0XVpl9raKp5%2Fr9TTbvHwLlwg9bz62OYasHjU6CuErUEKLHqucKYHEeuDCl5mwb5yY%2F1HA8bkX3II0QJRGCSmp9sIoug7XFxyCHtnea8BUR1w074r0quhyMQLnDmY7AWo7He1V0VdFc%2BN5VdyGKlOF%2FFKTXr%2BjWICEQIv4gNqYtkWH06Sq6DEOLfwCZPEexlmGvUsQ4lMsHj1XR1WUBhrpGM6f3aL5wRRcAjWM3uPPwmWOT4H1az7ax7YHaqLHvl7et6DKYwlmnDYhlj1H9S41QDO29FN8vV6vn0ofdi%2Br4rVFHIpqWr2o%2FMEq8H46Xc9uUagvoWGFjn647sHbQxyzKudJdTi3nypgiFp1Wzj3BiOPl3DOMOFzOPd8IwSqa0Vc597Hbc12oa6422pMadMWr91p0Qzq7RzekD9Gki2iOSQpkJXqX6dKtbXcB8tJsmrOLWTVD1HhcqFFXRxuESQahklwfvVO3iL%2BiDGQFs2QRLOIut%2FNVenVJ1h1OdwTcF2zWlZ3QJiJaUxjLAVfs1sXyooxBqTp1CmMYfXXrlkCJch1BW8PRVy1jnMolhQfv2FUSup4tmfMpOnbj0E3uPPwWTP%2BlSsy2sX2jx3jtjt0a023SBB%2BAZk9p4qcao8h48FN07HLnJ3JChTtS0bS8RTiC0eIdcWbHbuwZThrQ6diN%2Bo7dqO%2FYje%2FRsXuK6BU6doeNOLNjd9yIEzt2zzJCsIpm9B8%3D

things like this seem to be smaller with regular compression

because cjxl cheats

and even smaller <:CatSmile:805382488293244929> (took a few hours to compress)

Gotta make it smaller >:D

You again ! xD I did not search many many things for the encode, so yeah definitely there was a way to do smaller than me I used cjxl 0.8.1 and `cjxl -e 10 -d 0 --brotli_effort=11 --allow_expert_options --num_threads=0 -I 100 -g 3 -E 11` Apparently my output contains alpha (yours don't) and you used relative rendering intent <:FeelsReadingMan:808827102278451241>

https://tenor.com/view/terry-crews-its-me-surprise-hello-hey-gif-4532598

~~Discord embed moment™️~~

The alpha layer is probably why mine's smaller <:KekDog:805390049033191445>

yeah the only things idk anything about with JXL is metadata, alpha and channels/color space

**\_wb\_** _“Moore's Moiré”_ 2023 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKz0wB0qaGRlweqZnpGSUKhgZGJlyhBcWJuQU5qQqGXMGFpampValcQc4hCiZcXJlpCuHafrp%2B4Qp2CrqGZlwKCroKjmXpftpAEV1DGDccxDU25AIA

**\_wb\_** _“Pink pieces”_ 2023 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=HYmxCoMwFEX39xV3Lw9MrMFJsF2cHKqSWfTVBCIEGvr9fXW655z7iGWXXAJa8nHXtY2jQeIRyoVL%2FqxnTqJ%2Fn3JY6fWcYSxRfGNDh%2BoP%2Fjby6NXY3Qlg9N9DG9i4SydJshWYytbE8Np%2F

https://embed.moe/https://cdn.discordapp.com/attachments/824000991891554375/1089259058910998548/art_15.jxl

I like this guy in the middle

Cute, right?

That one is going for a hug

Jpegxl-cutout

https://embed.moe/https://cdn.discordapp.com/attachments/824000991891554375/1089395681715822633/hug.jxl

The image dimension parameters are a bit strange in Upscale mode, ```Width 902 Height 902 FramePos -870 -870``` Unfortunately jxl-art.surma doesn't render it correctly

**\_wb\_** _“Blue (da ba dee)”_ 2023 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOYKz0wB0kamZlweqZnpGSVgZmhBcWJuQU6qgglXkHMIUBVXZppCuLafrl%2B4gp2CriGXgoKugmNZOlAIwgNKJwNlQEyQVHBqTmpyiYIFKtecCwA%3D

hmm, I've tried to get some examples to work and I think my local jxl_from_tree is broken (got it from apt-get). For example: ``` if c > 1 - Set 255 - Set 0 ``` This renders as green. Hexdump of the resultant file: ``` 00000000 ff 0a fa 1f 01 91 08 06 01 00 30 00 4b 38 69 08 |..........0.K8i.| 00000010 c9 0d f7 07 48 70 00 00 |....Hp..| ``` Or the classic rule 110 (https://www.pouet.net/topic.php?which=12091&page=1#c568703) renders fine, but in black and yellow... What could be the issue?

Are you on latest git version? There was a bug previously where it would use a random RCT iirc

Try adding RCT 0

woot, that fixed it! Cheers!

**\_wb\_** _“Misty Glass Towers”_ 2023 image/jxl 40 bytes https://jxl-art.surma.technology/?zcode=XYw9C8JAEET7%2FRXTh4XcxWtTKILVIX6wtSSrOYhw6JHf70Ylgt28mcesU%2Bk1lwGuJkm9heA87TTdhmKdX9E5Py%2F3PCo8HTYnBKJ0RYcWNQEW9w%2BdvgQwosD9IruPJFVkw9bel3ULbvwbZoFFzKiiyHI2W0cdtStg%2F98EegE%3D

**\_wb\_** _“Industrial Revolution ”_ 2023 image/jxl 34 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOEKz0wB0pZmBlweqZnpGSUKpiYGXKEFxYm5BTmpQPkg5xAFAy6uzDSFcG0%2FXb9wBTsFQ1MuBQWgQCWQbWRqBuQoKOgqBKfmpCaXKOgawgT8XKEckF7d8HCgZm2%2FcJABBmhajFH5JlwA

**\_wb\_** _“Selection Bias (try changing that last constant from 133 to other values)”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKLShOzC3ISVUw4gpyDlEwNOXiykxTCNf20%2FULV7BTMDI15VJQ0FUITs1JTS5R0DU0sgDykVUYAPkoKlD5hsbGXAA%3D

**\_wb\_** _“Unexpected result from just NW”_ 2023 image/jxl 30 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMDTgCi0oTswtyElVMOQKcg5RMOAKz0wBypibWXB5pGamZ5SAmVyZaQrh2n66fuEKdgpGpqZcCgq6CkCOrgWQhSxnAOTD5AwRbFNDQy4A

**\_wb\_** _“Gradient 444”_ 2023 image/jxl 34 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMDTgCnIOUTAy4gotKE7MLchJVTDi4spMU0hWsFMw5FJQADIrgUwDIFNBQVfBDywIYelC5cO1%2FXT9wpEUOZal%2B2n7uSroGkEF3IsSUzJT80oUTExMuAA%3D

**\_wb\_** _“The fabric of reality”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=LYqxCoNAEAX79xWvl4UzqJDGImlSXRENWwfc6EECJ14av947sBtm5hbSZDEtbPG8j6wdXnF7%2F%2BLX2GBY%2F2a7QcNUjsbhYWFeEq%2BdA8KHWnnxyp4dSOFgidJmLElUc6u8lu6yLIdS6hM9LzgA

**\_wb\_** _“Nebulosity”_ 2023 image/jxl 32 bytes https://jxl-art.surma.technology/?zcode=XYoxDsIwEAT7fcX2kSWHhCAaCiigckFAV0fyQU4CZMA0vB67pRvNzNZy1JRnDjjuTuxwTu%2Fpnm7KHmKxhGXvcVC7zpnrwWN8flS%2FCtiF0gQXhBu2HUjHUTNduyhcoxMptQlSD19kXfavKZo%2B6vdvVvgB

**\_wb\_** _“Structure (1)”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOAKLShOzC3ISVUw4grPTAFKGBoYmXB5pGamZ5RA2FyZaQrh2n66fuEKdgrGhlwKCroKwak5qcklCrqGRhZgPlDOjAsA

**\_wb\_** _“Structure (2)”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOAKLShOzC3ISVUw4grPTAFKGBoYmXB5pGamZ5RA2FyZaQrh2n66fuEKdgrGhlwKCroKwak5qcklCrqGRhZgPlDOlAsA

**\_wb\_** _“Structure (3)”_ 2023 image/jxl 27 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOAKLShOzC3ISVUw4grPTAFKGBoYmXB5pGamZ5RA2FyZaQrh2n66fuEKdgrGhlwKCroKwak5qcklCrqGRhZgPlDOmAsA

**\_wb\_** _“Structure (4)”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOAKLShOzC3ISVUw4grPTAFKGBoYmXB5pGamZ5RA2FyZaQrh2n66fuEKdgrGhlwKCroKwak5qcklCrqGRhZgPlDOkgsA

**altzeitraub** _“eco-anarchism (green anarchism) square flag”_ 2023 image/jxl 33 bytes https://jxl-art.surma.technology/?zcode=8y%2FKTM0rSSzJzM9TMObKTFNIVrBTMOBSUNBV8FMwNLIAsoCCFVBBkHA4nBWcWqKga2FqYGJkacLFBQA%3D

The downloaded JXL for that one is completely green <:SadOrange:806131742636507177> . The PNG is fine.

I suppose this one is going wildly out of range, and things depend on how color management is dealing with that

the N 128 means that the G and B pixel values are 1024*128 at that diagonal (while the nominal range is 0..255)

while the R pixel values are -8504294, which obviously is also quite a bit outside the nominal range

it's not really well-defined how to render pixels that are claimed to be sRGB but they're also very far out of range. Likely different CMS libraries will behave differently on it, and things will be wildly different if you first clamp to uint8 or uint16 range and then do color management, versus doing things in float32

I see, thanks. So that means that currently one can't rely on the preview (even if a JXL extension is installed, but looks like it's not rendered as JXL then) and should ideally learn a bit better what the values are for. Thanks for pointing out the mistakes and for the clarification in general.

I think the jxl-art tool converts to 8-bit RGB so it clamps first and then does color management, which is similar to what you would get when running djxl on the image to convert it to PNG, but can be different from how it is rendered when no clamping to the nominal range is done

in general it's not well-defined how to render colors that are very much out of range — it's interesting and a bit puzzling that you're getting the image you're getting

Having everything looking like cyan (RGB #00FFFF) would be what I would expect for that image

somehow the very negative red removes the G though, until G itself becomes very positive

Somehow still a bit wrong (at least the preview looks different), but this would at least generate this image in JXL: ``` if c > 0 if W+N-NW > 32768 - Gradient -65472 - Gradient 32 - Set 0 - Set 0 ```

note that in principle, level 5 bitstreams (which includes all 'naked codestreams' like what jxl_from_tree and the jxlart website produces) are not allowed to require more than int16_t to represent modular pixel buffers, so something like the above is technically not a conforming bitstream

(libjxl uses int32_t internally anyway at the moment, but in principle implementations could use int16_t and that will certainly give different results in cases like this)

So max 32767 for W+N-NW and Gradient or do I need to calculate this different?

I saw a gradient with a way higher negative value in this channels and thought it would be fine

it's not just the values that occur in the tree, it's the actual sample values the tree produces

something like `- W 20` is technically already a problem since that creates a diagonal gradient with a value > 40000 at the lower right corner

so in an implementation that uses int32_t like libjxl, the bottom right corner will just look white, while in an implementation that uses int16_t (which is allowed by the spec for level 5 jxl bitstreams) will get an overflow and the bottom right corner will look black since it will have a negative pixel value

Is there some kind of writeup (going beyond https://jxl-art.surma.technology/wtf.html) that helps to understand how to have a conforming bitstream / how to stay within int16_t limits so that values don't overflow?

I don't think so. I suppose you could make a version of djxl that checks for out-of-bounds conditions and tells you if a bitstream is valid for level 5, level 10, or not at all

it currently doesn't check because having such a check would make decoding quite a bit slower (it's extra branching per pixel, which is not good for performance)

**altzeitraub** _“eco-anarchism (green anarchism) square flag”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=8y%2FKTM0rSSzJzM9TMOFyyixJSS0oyVAw5OLKTFNIVrBTMOBSUAAyK6FMMKcCzlFQ0FXwC0fmALVCWMGpJWBxGAsA

Should be fine now...

**\_wb\_** _“Eco-anarchist flag”_ 2023 image/jxl 28 bytes https://jxl-art.surma.technology/?zcode=8y%2FKTM0rSSzJzM9TMOFyyixJSS0oyVAw4uLKTFNIVrBTMORSUNBVCE4tUTAAsqCCICaYUwnngFT5hcM5QLkKJDlkM5D5RlzIcjAaAA%3D%3D

This one is actually green, not cyan 🙂

**\_wb\_** _“Eco-anarchist flag (another shade of green and one less byte)”_ 2023 image/jxl 27 bytes https://jxl-art.surma.technology/?zcode=8y%2FKTM0rSSzJzM9TMOFyyixJSS0oyVAw4gpyDgEKcGWmKVQq2CkYcCko6Cr4hYMZQLEKqBhINDi1BIVtyAUA

Ah! Looks like the color temperature program I use (redshift) is a bit too strong, heh.

Thanks for fixing it. Else I wouldn't have noticed it for a while since I barely use other devices than my PC.

It is definitely too strong... I need to keep that in mind when doing anything with colors in the future 🤔

Heh yes your screen looks like it has no blue at all

Looks like your computer is running on Argent energy

The French flag probably looks like the Belgian one on your screen

**\_wb\_** _“Drapeau Franco-Belge”_ 2023 image/jxl 35 bytes https://jxl-art.surma.technology/?zcode=C89MKclQMLQ04vJIzUzPKFEwNLLgcsosSUktAIqbc4UWFCfmFuSkKphwcWWmKVQo2AFVmHMpKAA5yUCOAZCpoKCrEJxagsI2NDBAKDIESwA5lXAdIHV%2BUAlU%2FVBbzIxR5CCWIqsFAA%3D%3D

For the french one it isn't that bad

But Luxemburg could almost be India

The other image appears darker due to the Discord theme I use.

I remember using a blue light filter to such degrees in the past, or at least close to it. At some point, you just stop noticing it, even if it seems really strong at first. I remember once going to the kitchen in the middle of the night (where my family usually keeps a white light on), and wondering why the light had suddenly turned blue, not realising I was just unused to the regular white light not being red. 😄

(Though I normally use light themes for everything, so the redness becomes even more apparent everywhere.)

that's white balance for you

depends on the environment

and why blue light filter is <:troll_face:905917618437910619>

the problem with doing it with redshift is that the naïve reduction in green and higher reduction in blue distorts hue and saturation

if you calibrate and profile your monitor with a warm white point, you get to keep relatively consistent colors

at some point, I prepared a couple of simulated images to illustrate the difference, I wonder if I still have that somewhere

if not, maybe I should prepare that again

maybe even write a blog post or something

(I would have to start a blog for that)

**altzeitraub** _“Clover”_ 2023 image/jxl 133 bytes https://jxl-art.surma.technology/?zcode=Cy7IycxLVTDgMqAIG1Kof6DtNzE04gJiBSMk2gxIm0Bp17yUYHBIcQWPBhhygIFoYGApmCHRJqMBholhAWOBRJtBA2w0heEIMFgAWUBTmsVolsSNTYEBA8QK5kDaHEpbmEFoS1BAmgFpM0iAGhoABUzNQAwjE6Sg5MpMU0hWsAOap6CgqxCcWoLCAgA%3D

**altzeitraub** _“Cute face with big eyes”_ 2023 image/jxl 372 bytes https://jxl-art.surma.technology/?zcode=7ZjBauQwDIbvegq9wCyWlMT2tVDovSx7LkzYDnTLQOf9qX45MzuHaaBNYfdgBjSOYyvhi%2BRf9t3htJ%2BPp2ce6Ofx7enP8WVmoV%2BHvXfZRA%2Fz4ffzCa3H48vhdabEm360k60Oxo0O0g%2Fd6sImlkI2sg5kxprJhE1IC5uRDmwjaQI1GcMKeuKmXydSTBFWc5xSvSU%2BiysNPFFmw3jB1ERS3aj5pWZWssTxxIyHi8SLjMvr3L%2Ful2%2B0%2FO3yVtT%2F3ME3fCtxOG4zcKE9AKxDEws7RE%2BM6QBvuijBLwUnBbMKU3AdUEtn96ELT1qHoxWcNJBpZKwvEbARjm1MB3gbYIroawAjBlUCXWSvavSkDnAteyXWOo21TpDDEzomz%2BPst8sli6Uz%2FFhCAhwgStiAKA7QxSRjQLmSmA5yVUrMKx03JYhGLEruovLpigYFo4dgPWNUYFxqm1bnXAKyo1zX56bDF00ez1rdZKfr87o%2BS6BTcmew2MR4c4JSZwwoV%2Frdl8aVSJS%2FlWJBIOZrmCHhsLaM7SBXKkYHVwGuRmwGQ18dl9qx7WUuMdmT%2Bwu7l4Y04rIutG%2Bvk5sPjIatDuw%2FIBnnRaxKI059Coxrt7aqMnS8lT9GcSDRdjc3T3U6T470djbuxoniYMxwMgZrYGoDmFpA9H6smE3hx3P0XnPd8eN88hvv

**altzeitraub** _“Cute face with closed eyes”_ 2023 image/jxl 195 bytes https://jxl-art.surma.technology/?zcode=7ZVNasQwDIX3OoUuMMWS7PxsC4Xuh9J1IaETmA6B5v6M9JJCF%2BnKhXYRAoqt2M%2Fmi%2BX3OC3DOC8XzvQyf759zNeRhV6nwVPW0PM4vV%2BWaJ3n63QbKXHVQyepFSiVAulBayWsYenICmsmM9aWTNiEtGMz0sxWSFNQk4IokcFH7yfSmCKs5jil95b4LO4pc0MtW4yXmJpIeg9q3tWWlSwxVmxjcRFspGzbeboN2z%2FaXqe2FvWfC%2FzCvxKH47ENXNHOAdahiSFmZDDmALgr0YFfAicNZn2ELvqA2h3sfpTwonU42gcnBTJFxfoVERHHcR1zANwHmHD6VoA4gypAh%2BpVRSbtAqx2qlwrYP%2BAH4yKVd143G66CO4uKqjlHk6DYjbCTbhi3bWTg2fwzMHGZZxoOLKFJUe0YGo5mBogej5cfS3x8lX037me%2BDwu%2FuEO

Attached is a image of the originals. So for some reason svg2jxl ignored the mouth and did something funny with the eyes but the results are still very cute (would say even cuter than the originals).

UwU

**MSLP** _“Spliney”_ 2023 image/jxl 139 bytes https://jxl-art.surma.technology/?zcode=7ZJBCsIwFET3%2FxRzgZZMbGOycSOCayt4ABttQEoXvT%2BmaSsibqSbLiQhzMAwfN7PJdR9A71xcvTh3vTQtHLan6FEqu4RWi9aKSw6q2j4Je1eivMdKoyCiS8Bw2SsHNp6ovSn9UGLRcLFYuSV7DdgmV467BoamLNEbrmoJP6piEkY10czEhz8G7NwwxW7GJ0EJUPle3DrZmX1pEonTw%3D%3D

**\_wb\_** _“Pharmaceutical Abstraction”_ 2023 image/jxl 31 bytes https://jxl-art.surma.technology/?zcode=C3IOUTA05XLKLElJLSjJULDgCi0oTswtyElVMOLiykxTCChKLXNMKnYtKlKwUzDgUlDQVXAsSw%2FX9lPQNQTygCqAbF2%2FcKgsSD44NSc1uURB1xTKB6r30%2FZzVTA0suACAA%3D%3D

That image can be generated with code like ```c image[y][x] = x & y & 1023 ? 1 : 0; ```

Sure, but the cool thing about JXL art is its consequential emergence from practical coding tools in a standardized image codec.

In particular, you don't get any programming language in JXL — you cannot write/execute arbitrary code, that would be bad for security and it's not something you want in an image format — and all you get is one iteration per channel, in raster order, over all the pixels, where you don't get to execute an arbitrary code expression per pixel but only a big nested if with a restricted language of conditions and a restricted set of return value functions.

So it's way more constrained than something like gpu shaders, but still it's expressive enough to do interesting things.

https://embed.moe/https://cdn.discordapp.com/attachments/824000991891554375/1101238767475495083/99th_JPEG_meeting.jxl

feels Squeezed Man

do you have the jxl_from_tree for this?

Yes. I used svg2jxl with some manual tweaks to its output

ah ok

(to add in a funky background)

**\_wb\_** _“Emergent Behavior”_ 2023 image/jxl 25 bytes https://jxl-art.surma.technology/?zcode=C3IOUTDgcsosSUktKMlQMOYKLShOzC3ISVWw4ArPTAEKGRpZcHmkZqZnlCgYW5hwcWWmKYRr%2B%2Bn6hSvYAXUqKOgqAJm6JhCWq4I5FwA%3D

https://jxl-art.surma.technology/?zcode=C3IOUTDmcsosSUktKMlQMOEKLShOzC3ISVWw4AouLE1NrUrlCs9MAUoZGxhweaRmpmeUgJlcmWkK4dp-un7hCnYKBlwKCroKQKauIZgVHg40FAA

I like these blobby shapes the default 8x upsampling is producing

**EV** _“Grid 1x1”_ 2023 image/jxl 26 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMODiykxT8FOwA7IUFHQVwhV0jUxNwUw%2FBRALAA%3D%3D

**EV** _“1px lines”_ 2023 image/jxl 31 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQsOAKcg5RMOAKz0wB8sxMuTxSM9MzSkAst6LE3NSA%2FGIFXUMg4uLKTFMI1w0PV7ADKldQ0FXwU9A1MjWDMkEsAA%3D%3D

If you just use set, it's smaller.

Also, it's funny that e10 gets 53 bytes.

**\_wb\_** _“1px vertical lines (simplified)”_ 2023 image/jxl 23 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOQKcg5RMOAKz0wB8sxMuDxSM9MzSkAsrsw0hQoFO6CkggKQGQ5lKijoKgSnlqCwDbkQogA%3D

**\_wb\_** _“1px diagonal lines”_ 2023 image/jxl 22 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOQKcg5RMOAKz0wB8sxMuDxSM9MzSkAsrsw0hXAFOwVdQy4FBV0gE8oITi0BagMA

Diagonal lines require 2px spacing, otherwise it is just a checkerboard pattern

Last line should be `- Set 1`

one could get away with either using `|W| > 0`

w8, sorry, I thought you were talking size-wise using embed.moe jxl-from-tree; but I mixed it up. What's the reason for `- Set 1 ` on last line?

I guess to make it look identical to this

ah, indeed

**\_wb\_** _“Barcode synthesis”_ 2023 image/jxl 25 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOMKcg5RMOAKz0wB8gyNLLg8UjPTM0rAzNCC4sTcgpxUBQsursw0hXAFO6BKBQVdBfeixJTM1LwSBV1DMD8crCc1RcGICwA%3D

Is there a way to extract the prediction tree from a compressed jxl image? (reverse the process on https://jxl-art.surma.technology/)

are you participating in the gCTF? xD

yes

there's a PrintTree function in libjxl, you can patch `djxl` to print out the tree

but I shouldn't give you more information 😉

okay thanks a lot!

https://capturetheflag.withgoogle.com/challenges/rev-jxl

for the curious, none of the jxl core dev was involved in creating that challenge (I was certainly very surprised when I found out it existed :P)

oh that's quite interesting challenge

My immediate thought was "Could've made that into bytes rather than KB"

And thinking about it, if they use pixel data rather than the raw data, there's nothing stopping it...

uh jxl-oxide doesn't match libjxl for that image

another bug for jxl-oxide

no surprises there, IIRC it did a bunch of weird stuff

I guess blending is buggy

Now that the ctf is over, could you please point me to the jxl tree documentation?

would probably make the challenge easier to reverse 😄

here is a tree_from_jxl tool I hacked up during the ctf, in case it is helpful to anyone it did require other hacks, like unprivating some stuff from libjxl and modifying DecodeFrame to add out param for the tree

hey, I'm the author of the challenge 🙂

if you're curious (spoilers!), here's the solution/writeup of the picture: https://github.com/google/google-ctf/tree/master/2023/rev-jxl/solution

or my short note about the solution: https://github.com/google/google-ctf/blob/master/2023/rev-jxl/challenge/solution.md

Sorry if has been already covered... Am I correct there is no modulo function with predictors? i.e. I want to draw a pixel every 8th column `x%8?0:1`... 🤔

this can't be done with JXL art as far as I'm aware

since the residues are all 0

you could probably accomplish something like this with patches, maybe?

or with lz77 it would be quite effective too

we should add some syntax to jxl_from_tree to encode some repeating pattern of residues, any forever repeating pattern should be very cheap with lz77

what you can do now is use a small group size and do it the exhaustive way (`if x > 7 if x > 8 if x > 15 if x > 16 .... - Set 1 - Set 0 - Set 1 - Set 0` or something like that)

**_wb_** _“or you can do this”_ 2023 image/jxl 25 bytes https://jxl-art.surma.technology/?zcode=c8osSUktKMlQMOQKcg5RMODiykxTqFCwA7IUFIDMcCBT1wLIVlDQBXJ0geog7ODUEjAbwjLgAgA%3D

taking advantage of clamping, the pixel values go like 1 0 -1 -2 -3 - 4 -5 -6 -7 -8 1 0 -1 ....

I guess I have an off-by-one there if you want % 8 but you get the idea

Thanks <@853026420792360980> & <@794205442175402004> ... was hoping to do some repeating pixel art... but realising this will be very hard 😕

I'm still intrigued by the idea of packing 8x8 patterns into a palette... but not quite ready to hand-craft a jxl file byte-by-byte (yet...)

https://discord.com/channels/794206087879852103/794206087879852106/1117167626586693652

If you'd be repeating the same symbol over and over one after another on the same page, like "¥ ¥ ¥ ¥ ¥ ¥" then patches should already do 90% of the work, but of course that last 10% can only really be gained by hand

how does the `jxl_from_tree` thing work? how does an image format have a programming language in it?

It's not a programming language. It's an expressive way of signaling a context model + predictor to use for entropy coding.

Think of it like the old huffman compression examples, a tree that goes down a path depending of what's before it. JXL just has more way of making those paths or jumping over them

https://jxl-art.surma.technology/wtf.html

**_wb_** _“Shalom, Salam”_ 2023 image/jxl 203 bytes https://jxl-art.surma.technology/?zcode=pVRNT%2BMwEL37V8wZqYtn%2FBHnwgGExAmtlpU4F5pCpNIgEQn49ziuE2J7QpGqHpx5L%2FPGM%2FOay7bfNK%2F9M2jx7%2Bo%2FSHHfbnyEkrS4adqn5%2F7wLM7P4G%2Bzfmzg7fPlodvB2bm4e921%2B0asSMJJP7HCkxXkH3OihD4xv9I1aOlPg%2BF05GOlhLMEBinEpkbPS7B%2BorbCEFvlAj%2FEw%2Fu5zvD%2BeFpSU%2F5wGsR4%2BnxrwLjan9rryoJ3SgfeKcPyY53r%2FSYudtj5Vffy0O7XfdvtodtC%2F97Bdrd%2Behv2327hEy586wDxcdADmEJb2xDOgCoCE1RVOEEzkGbgBDvCBJ4RlBEAK7i9DncrMj58BtZ1QQXyMXa0SCJLDvXumh7Uj6wTyxxXlO2BAWNTpGvBSaM5JpEBUU95mx3pPxZA8Ysx8hMqu08uMwvisknK1FUkMXcVGVm6igxyrlJkeFcpspyr7pddVUr9sIRFOQYcF2yk%2BKUp2cUc8fiyhXmLFhfNgPHWMn2nmEaSNgUx%2B4DPL5B1cKAoiVCPX6fwCZoMYOZGqqxLjVTT9x%2B0uCYzaWbG7HS5uaYTTapllbIqRYVc%2FVt5Uv0C

current jxl_from_tree gives a smaller file than the web version:

Huh, interesting

Will web version be updated?

**Anonymous** _“Unnamed piece”_ 2023 image/jxl 74 bytes https://jxl-art.surma.technology/?zcode=bZDBCsIwEETv%2BYq9l0DSNG1PgoroKYIKOYvGNpcqEkT%2F3qRp04b0tvPY3WFmo81dvUwLNTq%2BterM1ehnBxU6bS9AEdIP%2BMEKKCcIwIqbFW4EwCCGyeKvxZx42QO5Pzg0ALe9%2FjQiExIyyBO6sxObjt23oiLxmnTHuA4wmLCAIptiATuf0gfpU7FyDDDlCpLOrXzA2OqsDOCcJqzg6WW8ZYNAeudLXmzU151FhQr3hI8BpN3H4YG1ULppjboDpmhuwnKC%2Fg%3D%3D

I'm not convinced it isn't a programming language. It seems like it could even be turing complete.

Though I just started learning programming in jxl. ``` if [property] > [value:int] (THEN BRANCH) (ELSE BRANCH) ``` Can [value:int] be negative? How does it work with Bitdepth? If I have Bitdepth 1, can value still be negative?

it can't be turing complete because it walks down the tree once for each pixel which is a finite number of times, and always terminates, no matter what

turing machines have the possibility of not terminating

Yes, I agree. But computers aren't turing complete either if we are being strict.

but that's a limitation of hardware, not a limitation of language design

the value can be negative, yes for example W-WW is a property that takes the value to the west and subtracts the value west of that. this difference can be negative as can the thing to compare it to