JPEG XL

<https://gitlab.com/wg1/jpeg-xl/-/blob/master/plugins/gimp/file-jxl.cc#L60-62> this is.. different too?

<https://gitlab.com/wg1/jpeg-xl/-/blob/master/plugins/mime/image-jxl.xml#L8-9> this seems to be the same as fingerprint used in squoosh

That is correct

That is an ancient version of the signature

We had some 4-byte signature a long time. It looked like ×LM\n, and what we were making was called XLM, the jpeg xl (validation) model - like HM is the validation model for HEVC.

That was before we learned that JPEG likes to have its markers as two bytes, 0xFF followed by something else. 0xFF0A is the marker for "start of jpeg xl codestream" and we don't have any other markers

Then there was Brunsli, which we at some point wanted to consider as a valid jxl, but it already had its own signature. That's probably still in some of that old code, but it's no longer valid - Brunsli is not valid jxl.

Then there is the isobmff container which can be used to add metadata. The payload is a `jxlc` box that contains a jxl codestream, so you can strip the rest if you want and just keep the contents of that box.

And then there's the old JPEG, which we at some point also wanted to consider valid jxl, so libjxl would be literally a replacement for libjpeg since it would also read old jpeg. But we didn't go that road after all, because the libjpeg api is a mess and we cannot really be a drop-in replacement anyway, applications need to be prepared for an alpha channel or animation anyway so it is better to make something new.

So there have been 4 signatures during development, but only 2 made it to the end and are now valid ways to start a jxl: the naked codestream that starts with 0xFF0A, and the container that is isobmff with a `JXL ` signature (ftyp `jxl `, payload is in the `jxlc` box).

Ok we are getting way too <#794206087879852106> here

https://c.tenor.com/syaBQDPmW2EAAAAM/change-subject.gif

https://tenor.com/view/off-topic-discord-brick-hill-gif-19084558

meh, for internet video it's still enough

for movies it's a sin

I watch everything in 480p for the most part since my internet is horrific

<:AV1stonks:781612882176507915>

i think youtube has hurt people's perception of 720p

since their bitrate is so abysmal

1080p too

youtube gives me some pretty bad banding in the blacks

not sure if it's because I have a 500 nit laptop...

Ah yes, time for the classic. https://www.youtube.com/watch?v=h9j89L8eQQk

I think it's mostly because YCbCr effectively deletes the lsb (least significant bit) of R and B, then lack of chroma DC precision deletes some more lsbs, and the transfer curves that are in common use for SDR were designed for the 80-nit CRTs of the 1980s, so even without losing any lsbs, 8-bit RGB is not really enough anymore for smooth gradients on current screens (which are brighter, more accurate, and don't get the natural horizontal smoothing of an analog CRT).

blu-ray gets around this by using absurdly high bitrates: https://x0.at/WZa.mp4

enough for the noise to remain and act as dither

That was a very stupid show, I'd watch another 8 seasons of it

Happy to see that guy on Umbrella academy now

these days, I’m rewatching the first three seasons (3rd watchthrough, iirc)

it’s really great

Don't get null as a license plate

my new license plate: > X5O!P%@AP[4\PZX54(P^)7CC)7}$EICAR-STANDARD-ANTIVIRUS-TEST-FILE!$H+H*

OK, what exactly does the "this" emoticon mean? I see people use it a lot here…

It means you support the above comment or that the above comment is correct

like <:This:805404376658739230>

thanks 🙂

Needs Moiré pattern

for %i in (C:\Users\User\Documents\f\*jpg) do cjxl "%i" "%i.jxl" -j -s 4 -I 4.44 -m --mquality=23 -s 7 --num_threads=2

try this command with 0.3.3 if you want to use modular as low as you can

5,5 mpx in 45 kb

Oh

I'm not top 10

😦

Super sad.

Don't worry, you can get better quickly.

It's quite easy in fact.

Just spam questions about AV1, Opus and JXL. <:Stonks:806137886726553651>

it's not THAT jpeg'ed when it becomes gray

Does anyone else use ECT? One of the options is `--strict` which is said to "Enable strict losslessness". Does ECT discard data by default? Like when just using `-9` on jpg or png files.

Didn't think it did.

<https://encode.su/threads/2274-ECT-an-file-optimizer-with-fast-zopfli-like-deflate-compression?p=53737&viewfull=1#post53737>

So it doesn't affect image data. Thank you for clearing that up, Scope.

It uses zopflipng and jpegtran, which are both lossless. I feel pretty stupid right now.

Zopfli only as a base, ECT has had a lot of changes and optimizations can be lossy, for example Pingo converts all PNGs to 8 bits per channel

1 BILLION VIEWS!

https://www.youtube.com/watch?v=CSvFpBOe8eY

4k AV1 is available

https://www.youtube.com/watch?v=9VvmixeowNI

<:Thonk:805904896879493180> https://status.gitlab.com/

Oof

> `DROP DATABASE TABLICE` Greetings from Poland 😉 Tablice (rejestracyjne) 🇵🇱 = (license) plates 🇺🇸

the tipsitaliani article is not bad but this way better

more simple

no i don't use the fonts i make

https://tenor.com/view/reaction-my-eyes-cant-unsee-burn-gif-7225082

this is an excellent font for ransom letters

i read dm sans is good for smaller px like 6 px

and is also readable

6 px is still too low

is designed maybe for 8px

And threats of violence, it mirrors the 'letters cut from magazines' aesthetic well.

Bee boop

some of my favorite music is somewhat blip boopy

Chiptunes?

oh, I actually wasn’t thinking of that, but indeed yes

I was thinking of electronic music from the 70s

the Berlin school of electronic music

I like anything from 3-channel C64 SID music to 4-channel Amiga MODs 😅

https://youtu.be/HiQrUDIah4E?t=29

Berlin school is blip boopy but in a rather organic way, I would say

this is where the genre started: https://www.youtube.com/watch?v=EhQpXD2Z9WQ

recorded at the end of 1973

I just realized that in a way, English is little-endian while French is big-endian

If you consider the substantives the most significant bits of language and adjectives the least significant bits, that is

the rules about this in French are quite complicated, I don’t know if I could summarize them verbally

sometimes the adjective must go first, sometimes it must go after the noun, sometimes it can be either (but then it can also subtly change the meaning)

a big cat: _un gros chat_ a huge cat: _un chat géant_

or _un chat énorme_, or _un énorme chat_ (this one works in either position)

I know, but besides a few simple adjectives like big/small/young/old, most of them go after the noun, no?

true, probably most of them

outside of a few set phrases maybe

for example, North Atlantic Treaty Organization (NATO) is a quite little-endian way to say that you have an organization that is based on a treaty which is between countries around the Atlantic, specifically the North part of it

_sale histoire_ (“sad story”) vs. _histoire sale_ (“dirty story”)

in French it's OTAN, Organisation du traité de l'Atlantique nord

_sale_ is literally “dirty” but _sale histoire_ has become its own set phrase

which is exactly flipped and feels more big-endian

I was going to try it with “JPEG” but I’m having trouble fitting in the J

seems that it would be the _groupe conjoint d’experts photographiques_

GCEP

it sounds redundant to say that the group is joint, though

isn’t it enough to say that it’s a group?

in which case it would be PEG in English and GEP in French

in english it's becuase it's joint between iso and iec

no

everything in sc29 is joint between iso and iec

the 'joint' in JPEG is because it's joint between ISO/IEC and ITU 🙂

the 'joint' in JTC1 is because it's joint between ISO and IEC

“of” in English can be used to break the order

so much bureaucracy

for example, I studied at the _Institut National des Sciences Appliquées_ (INSA), which could be the “Applied Science National Institute” (ASNI) or “National Institute of Applied Science” (NIAS)

this example kind of shows how much vocabulary those languages share

National Applied Science Institute would sound best to me, except that abbreviates in a rather unfortunate way

ISO took a fun path: - in English it is International Organization for Standardization (so that would be IOS) - in French it is Organisation internationale de normalisation (so that would be OIN) The compromise was to abbreviate it to ISO 🙂

The only way to make everyone happy is to make no one happy for that it seems

there is actually evidence for that

> This simple thought experiment is consistent with the hypothesis that, within limits, people appear willing to sacrifice absolute consumption for improved relative position. The same hypothesis has received strong support in the empirical literature on human happiness and well-being.⁵ https://web.archive.org/web/20121221184110/http://www.brookings.edu/gs/events/externalities.pdf

how's that unfortunate? 😄 in italian it becomes "noses"

https://media.tenor.com/images/42c73fb75be778b3e83a7e334d9d1af3/tenor.gif

ah

in english it's almost always spelt with a z, I didn't make the connection

same in French but they would be pronounced the same way, i.e. /nazi/

since a single ‘s’ between two vowels is pronounced /z/

case -> /kaz/, casse -> /kas/

https://youtu.be/sWblpsLZ-O8

Btw, smaller than 1x1 pixel AVIF

That is an impressive one — the graphics are mostly epilepsy stuff, but even just the music is pretty incredible

Designed Glith-Art

shoulda been Designer Glitch-Art, then you could have charged more for it

Maybe we need to do 256 byte jxl bitstream demos

Make an art gallery full of 256 byte jxls

"What can we do in the size of an avif header?"

E.g. a nice 1024x1024 pixel sierpinski triangle should be easy to fit in that

Maybe someone can come up with beautiful splines that happen to compress well?

Or something made out of 256x256 dcts with only a few well-chosen nonzero coeffs, that could also look quite nice potentially

just try decoding random 256 byte bitstreams (maybe with non-random header)?

and checksums disabled

https://i.redd.it/nikixial60s21.png

https://i.redd.it/sftlxd5w8nj41.png

Could try something like that, a bit tricky to make valid bitstreams that way though (most would be invalid because of premature eof or invalid ans endstate)

Yeah, FLIF was nicer in that regard.

Also most of them would be ugly, I think 😆

Fair.

Flif had near perfect avoidance of redundancy, I remember doing a kind of 'redundancy fuzzing' where I would look for bitstreams that didn't decode to find out what hidden redundancy we weren't exploiting yet

That's cool, but not so great for speed, things like avoiding writing bits that can in principle be known from certain constraints is not the best strategy to make branch-free code 😅

I think the easiest way to make funny jxl bitstreams is to hack the encoder directly and make it write interesting hardcoded things.

Make a mock decoder that after EOF or invalid state just always returns the most likely symbol?

Then use that to sample modular trees that result in pretty pictures.

Maybe a MA tree human-readable syntax would be nice, and an encoder that takes a tree as input, with singleton histograms in every node so the tree description itself defines the whole image.

Histogram can just be the singleton 0 for all nodes. Nodes have a predictor, multiplier and additive constant, so don't really need any other histograms.

x and y are properties so you can set individual pixels or regions manually if needed

And in that case, there is literally 0 data needed to encode the image itself?

Yes

So just need to make a tree that fits in 220 bytes or so. Tree itself is compressed so if it has some structure that will help

It's an interesting class of quite expressive cellular automata, such MA tree-only images. When I retire, I am going to spend my days exploring that space 😂

he retired

just wondering, how large is the AVIF header?

300 bytes normally, although a tool like rav1e can create a 188 byte header.

oh

that's... kinda huge

how large is a JXL header? for comparison

Depends on the mode, and what you consider header vs data.

ah

it's pretty cool that you can fit sierpinski triangles in less than 256-byte JXL files

Airfryer bitterballen

In highschool biology lessons, we at some point did a test to find protein/meat in bitterballen. We failed to detect any.

There's supposed to be meat in them though <a:ThinkingSphere:821038590091329557>

was the test ever positive for anything else?

Yes, but it was a bit low tech and needed a high enough concentration of meat, and the bitterballen just were too diluted with water and starch and whatever it is they put in them.

> and whatever it is they put in them. <:monkaMega:809252622900789269>

why does it “hurt” to look at this

also, it seems to move on its own if staring at the center while sufficiently zoomed in

I suppose images like that are good at exploiting weaknesses and limitations of the human vision system

https://www.youtube.com/watch?v=3GLZAIaXudk

less than 100 bytes for a 2048x1024 sierpinski triangle

oh wow

Epic

PNG right after decoding with `djxl` was 39 KB, but with `optipng -strip all -o7 -zm1-9` I got it down to just 7 KB

5 418

How did you do that? Mine is 7 123. Is that the same JXL source that I mentioned?

ECT or Pingo or ECT+Pingo, Optipng is old, no longer the most efficient and very slow

Sierpinski is easy for png because of all the repetition

When I have time, I am going to make a beautiful image that is very small in jxl and at least 100x larger in anything else, even allowing lossy (if it still looks somewhat similar)

Thanks for those links! 😃

My comparison includes all popular PNG optimizers: <https://docs.google.com/spreadsheets/d/1ju4q1WkaXT7WoxZINmQpf4ElgMD2VMlqeDN2DuZ6yJ8/edit#gid=1741276444>

Is `Pingo + Brotli` used as some kind of performance anchor? Because I don't think you can use Brotli with PNG in a way that's compatible with current PNG viewers...

Not for viewers, but for websites it's possible when PNG is uncompressed

You're talking about `Content-Encoding: br`?

Yep

I don't know whether anyone actually uses them that way

Yes, but it is theoretically possible and interesting to compare

Honestly, the best PNG optimizer overall would have to be ECT.

Fast, efficient, open, multi-platform. It also has recursive search and parallel file threading. <:megapog:816773962884972565>

Beep boop

🌹 are red

🫐 are blu

I have no rhyme

Strawberry glue

Spam for the spamgod

<:Spam:716551944918925393>

I like her coral reef necklace 🦐

So, does anyone read Asterix here?

Not really read, but I watched a couple of movies.

I have it in Bärndütsch

(but mostly read it in French)

I read quite a few Asterixes when I was a kid. The translation in Dutch.

<@794205442175402004> Did you have one in Latin too?

Ah yes, I think so

I remember Asterix Apud Gothos.

i read them in english when i was young

I have a Harry Potter book in Latin somewhere 😄

(never actually tried to read it)

`\\\\\\\\\\ //////////` ` \\\\\\\\\\ ////////// ` ` \\\\\\\\\\ ////////// ` ` \\\\\\\\\\ ////////// ` ` \\\\\\\\\\ ////////// ` ` \\\\\\\\\ ///////// ` ` \\\\\\\ /////// ` ` \\\\\ ///// ` ` ///// \\\\\ ` ` /////////// \\\\\\\ ` ` ///////////// \\\\\\\\\ ` ` ////////////// \\\\\\\\\\ ` ` \\\\\\\\\\ ` ` ////////// ` ` ////////// \\\\\\\\\\ ` `////////// \\\\\\\\\\`

Let's make cjxl output that as a banner

Or maybe only when doing -v 😂

Are those bits at the bottom really needed? I think it might be better without those

Could also make the L symmetric with the J, I think

And still have a recognizable X

l.

``` \ / \/ /\ \__/ \__/ ```

Something like that also works. Could do the J and L in different colors

This is what I mean with the J: ``` / / / \__/ ```

Can animate it and let the L come from the right until they cross to make the X

``` \/ _/\_ ```

Or ``` \/ \/\/ ```

FLIF still holds its level quite well, just 1.82x of .jxl size 😃 `./flif -kNX1 sierpinski.png sierpinski.flif`

Out of interest, what settings did you use to encode this image? I've managed to re-encode the same image in 99 bytes (-s 9 -m -g 3), but I can't seem to ditch those last 2 bytes!

I don't remember, maybe `-P 0` ?

(99 byte version)

That did it, thanks!

It's pretty amazing that this can be stored in just 97 bytes, JXL is magic

it is a best case for the format

Sure, but it's still magic!

it find the rule for the next line of pixels and encode only the rule

here another nice small jxl image

each of these images fit in the header overhead of an AVIF <:kekw:808717074305122316>

I can kind of understand how the first two were done, but how on earth is the last one only 270 bytes‽

last one is also a cellular automaton, like sierpinski

I wrote a blog post on that back in the flif days: https://cloudinary.com/blog/compressing_cellular_automata

Why doesn't cjxl learn the "nearest-neighbor rule"?

This looks a bit like the logo of a golf club. 😁

because that's a rule that would be something like `if (x % 4 != 0) pixel = left; else if (y % 4 != 0) pixel = top; else pixel = whatever`

and MA trees cannot concisely represent `x % 4`

JPEG-XL beats FLIF by 3 bytes in encoding the rule 30 example image, pretty amazing!

(-s 9 -m -g 3 -I 3)

when I have time I want to write a program that takes a textual description of a tree and makes a small JXL image out of it 😄

Some recent posts in this thread seem like a training for making contributions to https://allrgb.com. 🙂

I converted https://allrgb.com/thingy (PNG, 56 086 B) with `-q 100 -s 9` and the result is quite impressive (2 956 B).

Have you tried adding `-g 3 -E 3`? It improves compression in majority of the cases.

I'll try…

Much better, thanks for the hint!

It also needs -I 1

I know about that switch, too. Sometimes it may help, sometimes it may not. `-g 3 -E 3` are IMHO the most "stable-working" compression enhancing parameters (they give the most consistent results, I've got yet to see an image that degrades with those switches).

-I 1 seems to knock off about 100b in the lower speed settings too. I suspect it only becomes relevant in cases like this where it is sub 1K and you're trying to squeeze every byte you can out of it.

This is not really off-topic

Off-topic for off-topic

https://c.tenor.com/lCRnrFscGW0AAAAM/john-cena-smile.gif

https://tenor.com/view/spit-take-laughing-lmao-gif-9271317

make this the new <#794206087879852106> ?

**March 23**, 1989 a world-ending asteroid flew right passed Earth. A distance the same as the diameter of the moon's orbit. Happy Near Miss Day!

That's like a JXL demo scene

Yes

100 byte jxl images

If only you could share the MA tree across frames in animated jxl.

ehhh

but you can make a multi-group image and copy parts of it to other frames, so close enough 😄

You can put all frames in an oversized referenceOnly frame and then have frames that are all just one big patch.

ok

<@!416586441058025472> can you give me your DNS or proxy setting?

or you can check it

is difficult

ok

All websites cannot be connected or just specific?

proxy keep default should be fine

ok

thanks

try tracert www.youtube.com

in windows cmd

I think that probably a ISP issue...

what it does

it said traccia completata

completed traces

35 ms

and how to block chrome CAPTCHAS

how lastrosade does

can we talk in av1 server

Avoid google CAPTCHAS, you need change your ip.

ok

35 ms isn't bad no?

it checked 10 times is it wrong?

is fine

google is very sensitive for now, I think covid bring too many network traffic to google server.

you can consider use Privacy Badger

If you keep getting Google CAPTCHAs then you likely have malware that is making a lot of automated requests from your IP address.

that's i was going to say first time but i didn't know how to explain it

Also check you router firmware

<@!416586441058025472> Don't take screenshots. Just download youtube video thumbnails using youtube-dl

where you found the screenshot of yt?

I don't understand

which text channel from there

https://discord.com/channels/794206087879852103/803645746661425173/824992146238210049

It's better to download thumbnails directly from youtube than taking screenshots.

Yes

I just lost a comrade I've known for 17 years due to a sudden brain tumor.

Life can be cruel.

oh no 😦 I am sorry

I feel very sorry for his wife and two kids. So sudden.

Good morning 🌅 and happy summer time

We're in the next decade and Poland (and EU in general) *still* uses daylight saving time, ugh...

I thought that 2020 would be the last year of clock fiddling, I hate this thing

US also does it, and they did that two weeks ago, which is always confusing when scheduling meetings because the usual time difference is off by an hour.

People invented time so it should serve them, not vice versa. I've been a long-term (as **L** in **XL** 🙂) advocate of abandoning the summer time.

Summer time is very usefull in sud Europe, you can use all day light

Gotta love how they were all like let's save fuel by changing time, because there's a fuel shortage... to then barely get a positive difference, but big backlashes <:kekw:758892021191934033>

That meme is already done <:FrogSupport:805394101528035328>

<@!228116142185512960> what's "Sharpness" slider in AVIF's advanced settings in Squoosh responsible for?

<@456226577798135808> I think <@!710762823986446367> might know more, but I think it mostly related to spatial noise shaping

Similar, yeah. Higher sharpness means more AC coefficients. It's pretty experimental.

But THAT one is definitely for <@!228116142185512960>: Squoosh reports compressed filesize `9.91 kB`, but after saving the file to disk Windows Explorer reports the filesize as `9.67 KB`. If we divide bigger filesize value by smaller one, we get 1.024819..., which is pretty close to 1.024, which is nothing else than `1024/1000`.

It's not AVIF's fault, I first discovered it when playing with JPEG XL in Squoosh.

https://www.npmjs.com/package/pretty-bytes - "Note that it uses base-10 (e.g. kilobyte)"

I think a kilobyte being 1000 bytes makes the most sense, in the end. I've had to fight my instincts on that one though.

I don't think I'll ever use the word kibibyte though

when powers of two make sense, there is nothing special about 2^10 anyway

2^8 and 2^16 are "rounder" numbers than 2^10

it's sufficiently close-ish to power-of-10 equivalents and sufficiently round for RAM makers to take advantage of (nice numbers without fractions usually) so I can see why it's prominent in computer related stuff despite not making much sense for stuff like file sizes

American here, metric bytes are stupid. it's all based on binary, just use 2's

“kilo” has a certain meaning, I don’t think we should change it just because it’s applied to bytes

and yeah, Windows Explorer reports sizes in powers of 2 but with SI prefixes (except with K instead of k)

the NIST agrees: > the US National Institute of Standards and Technology (NIST) requires that SI prefixes only be used in the decimal sense

It's only a matter of time before the robots take over and change the standard number system to hexadecimal or binary.

BTW <https://www.twitch.tv/revisionparty>

http://n-gate.com/ > The International Standards Organization attempts some damage control, but fails. Hackernews is mad about having to pay for access to standards, but some ISOs arrive to explain that they have to charge this money in order to pay for the accountants who manage the money they collect. Since there is no possible way for a committee of technical experts to collaborate without an army of bureaucrats charging for their work, everyone (both on Twitter and on "hacker" "news") continue to do nothing about the current situation.

https://www.keychron.com/

Holy crap, I had no idea that VHS has such low chroma resolution: > In modern-day digital terminology, NTSC VHS is roughly equivalent to 333×480 pixels luma and 40×480 chroma resolutions

(from https://en.wikipedia.org/wiki/VHS)

wow that explains a lot

VHS was designed to capture broadcast TV, which also had very limited chroma because they already had grayscale TV and added color later, literally in the spare bandwidth

yeah, they did some extreme math hacks to produce a colour signal that still worked well with a black-and-white TV

Very similar to how stereo audio was added to radio.

https://en.wikipedia.org/wiki/Colorplexer

The use of YCbCr420 is still in a way a remnant of those analog days

That's impressive, thanks for sharing.

Very strange weather today

It was just snowing while the sun was shining

And now it is hailing

And now it is snowing again

In April

Wtf

It was... snailing?

https://twitter.com/NuSan_fx/status/1378759959815585795

<@!794205442175402004> quick fix necessary ASAP pls, I'm live editing Polish wiki article and I need to link to a source, could you add some kind of `id="update"` to `<h2>UPDATE</h2` on https://flif.info for easier linking?

sure

if you want to make such updates, you can also make a pull request here: https://github.com/FLIF-hub/FLIF-hub.github.io

I did it myself now

http://flif.info/#update

BIG THANKS! 😃 I know I can make pull request, but it was necessary ASAP.

<@!794205442175402004> unfortunately plwiki admin found some problems with lack of sources to those paragraphs in the Polish article (here are their original English counterparts): > Prediction is run using a pixel-by-pixel decorrelator without side information, including a parametrized self-correcting weighted ensemble of predictors. Context modeling includes specialized static models and powerful meta-adaptive models that take local error into account, with a signalled tree structure and predictor selection per context. Entropy coding is LZ77-enabled and can use both Asymmetric Numeral Systems and Huffman coding (for low complexity encoders or for reducing overhead of short streams). and this one: > Animated (multi-frame) images do not perform advanced inter-frame prediction, though some rudimentary inter-frame coding tools are available: > - a frame can only update part of the canvas; > - a frame can not just replace the contents on the canvas, but also be blended, added or multiplied to it; > - up to four frames can be 'remembered' and referenced using the "patches" coding tool in later frames. Any ideas which publicly available sources can be used here to be linked to? Someone sooner or later will probably ask to put some sources on enwiki, too, because Polish version was translated from the English one.

good question... I'd say CD/whitepaper/SPIE open access paper but none of those have all the information IIRC

and the (F)DIS is not publicly available

Seems like `Template:Original research` will have to be involved... 🙁

https://techdator.net/microsoft-edge-browser-now-is-the-third-most-popular-desktop-browser/

I guess only the reference software is a complete, publicly available and reliable source at this point.

Very strange, this snow after Easter

April Snow is a thing here)

normally we get occasional snow from December to February. March is rare. April I don't think I've ever seen here.

it's snowing in Zurich too...

From Wikipedia: > up to four frames can be 'remembered' and referenced using the "patches" coding tool in later frames. https://gitlab.com/wg1/jpeg-xl/-/blob/100e3c7e/lib/jxl/common.h#L90 seems to differ... <:Thonk:805904896879493180>

ah, that's for patches, not for blending

although I guess how wikipedia is currently written is a problem

... don't see why I wrote 3 there though

Both patches and blending are mentioned in that paragraph

https://discord.com/channels/794206087879852103/806898911091753051/828758158008778803

^ the second one from there

First two points reference blending, the last one is about patches.

I think that 3 is actually wrong, but I'm too tired to figure it out today

By the way this is completely outdated, last commit to it will be 1 year old soon <:monkaMega:809252622900789269> https://gitlab.com/wg1/jpeg-xl/-/blob/master/doc/xl_overview.md

slightly outdated, yes 😆

> slightly

<:kekw:808717074305122316>

If someone updates the doc, this could be taken into account as well:

Levels of headings (some of them seem incorrect)

If there's the "Lossless" heading, there should also be the "Lossy" heading (they are equally important).

I am writing an experiment report and I almost wrote MEMORGY instead of MEMORY

<@456226577798135808> a little fucking warning would be nice

➡️ ➡️ ➡️ **WARNING:** WATCHING THE `COCA_COLA_ESPUMA_1.mp4` VIDEO ABOVE CRASHES DISCORD ⬅️ ⬅️ ⬅️

LMAO, 4:4:4 with extreme VFR and SVC scaling <:kekw:808717074305122316>

Doesn't work for me. <:PepeSad:815718285877444619>

Crashed on win 10 desktop app version

Doesn't crash android and linux app

Discord just in generally crash quite often

doesn't crash discord on ubuntu

pepelaugh

dumbasses tried

It depends on your decoder.

A software decoder shouldn't pose problem.

A hardware decoder however...

wait guys

have you seen the new discord code embed

can you see it?

yeah it's fuckin sick

looks like this for me

ooh even in compact mode

hope they bring it to mobile soon

wonder if it does the same for large text messages

nah but it should

for code blocks at least

it does offer to upload a text file if you paste more than 2000 chars into the message box since that's the message size limit

yeah exactly

<@!263300458888691714> i've had discord open for like 10 minutes in this chat and i've seen you start typing and stop

This was initially just a file download, but after a day or so the text was embedded. https://discord.com/channels/794206087879852103/824000991891554375/826399106623733802

are you having a anyreusm

sure

lol

I have now finished animating a dumb nerdy video about PNG https://mincerafter42.github.io/assets/png.mp4 voiced by my friend and written in probably the actual 1990s thanks to whoever it was in this server that gave me the link to colour correction stuff that let me implement it in Blender nodes

https://www.youtube.com/watch?v=kkilHvxUnjM

Time to go. 😛

continuation of https://discord.com/channels/794206087879852103/794206170445119489/831925128886485063 if there is more to say

When will 'the year of the linux desktop' be?

what would it take for there to be one?

It's something now often used satirically to show when linux becomes a major player on the desktop

If there's an architecture change from like x86 to ARM on desktops then I think Linux has an apt chance to shine, since software on windows and MacOS will be forced to go through a compatibility layer or get ported too

When chromeOS becomes the dominant operating system

Not that the MacOS compatibly layer is anything to scoff at

They played their cards well

right, but I have become somewhat desensitized to the joke so I approached it somewhat seriously instead :p

Linux seems to be growing in userbase

Not particularly fast

But Linux at least has it's niches

Wayland might hurt some of them

Yeah I've tried wayland

by currently making proper color management a pain if not even an impossibility

It seems like a good technology but there's just not enough improvements over X to make it worth it

(there is a huuge thread about this on pixls.us)

https://discuss.pixls.us/t/wayland-color-management/10804

Overall a meh kind of thing with flaws even compared to X

with a few interventions from Graeme Gill (author of ArgyllCMS) and Florian Höch (author of DisplayCAL)

you know what i don't understand

when the website is called rapidapi

and the average latency is legit 500ms

and the actual website itself, not where you make api calls is also

slow as fuck

That's just marketing. 🙂

are those updates in the software necessary for those who have a windows 10

and already can have a good experience like chrome with youtube bookmarks

and all the new thing required

like what is bad about chrome stable?

what are the percentual in 2021?

if you can link something

https://www.reddit.com/r/DataHoarder/comments/mr50yh/my_zfs_645tb_server_dual_intel_gold_5222_512_gb/

Ah yes, r/DataHoarder. I love them ❤️

https://www.hdblog.it/curiosita/articoli/n536359/streaming-video-inquinamento-ridurre-impronta/

source:

https://edition.cnn.com/videos/world/2021/03/24/video-streaming-environmental-impact-project-planet-lon-orig.cnn/video/playlists/project-planet/

https://www.tomshardware.com/news/hard-drive-ssd-shortages-imminent-if-new-cryptocurrency-blooms

> Unlike Bitcoin, which is based on proof of work, Chia utilizes a proof of space and time model. Chia reportedly arrives as an eco-friendly cryptocurrency. Bram Cohen, who's best known as the inventor of BitTorrent, created Chia to leverage the free space on storage devices. The basis behind Chia is that hard drives and SSDs use less power, are easier to come by and cheaper to purchase.

i think for preview q 37 should be used

it don't read all download this file it's a resume of what i deleted, i deleted some messages

Just use avif for thumbnail

How does one get the <@&833056294083493889> role?

Be in verbose mode

Ah, I had bot spam muted

Now I see

You froze Fabian 1 level from being expert <:kekw:808717074305122316> <:Spam:806628077656473650>

<:kekw:808717074305122316>

🦾 <:Poggers:805392625934663710> 🤳 🦵 🍗

<:kekw:808717074305122316>

I just realized that I had a dream involving AVIF last night