JPEG XL

could use google drive or something like fileditch

If speed doesn't matter, then try private telegram channel

It has limit of 2 GB per file

~~split~~

Exactly not enough capacity, unfortunately.

by individual jxl images <:FeelsAmazingMan:808826295768449054>

xD

Maybe I could blow the dust off my Raspberry Pi and host it there

it's ~20gb total?

I could put it on https://old.lucaversari.it/jpeg_xl_data/ 😛 (or bigbuckbunny.lucaversari.it if you like that better)

That would be lovely

https://bbb-jxl.lucaversari.it/1920x1080_24fps/

Just werks

You can even see how long it took

Would there be any way to check and see if that image has any VarDCT in it? If not, then we can assume it's almost assuredly not JXL, or if it is an incredibly poorly set up encode

it is jxl, just low quality, it comes from <https://boards.4chan.org/g/thread/98087793/jpeg-xl-09-released>

It both intruiges me and pains me reading that thread

4chan's /g/ board is incapable of productive discussion

> I want 3KB QR Codes > Just buy more hard drives and use TIFF > It didn't look good on quality 5 It's certainly a place

More specifically, that Discord post came 8 minutes after this. https://boards.4chan.org/g/thread/98203227#p98206338

I use a tool called rust-parallel myself which does need a bit of scripting, not sure if there is a gui for batch transcoding or not

Powershell 7 has some useful utilities for that IIRC, but that's something you have to download separately on Win 10

there's many options on linux if you want to use wsl

I use a tool called rust-parallel and sharkpd's `fd`, I then invoke the command like so in powershell ```ps fd --extension png | rust-parallel.exe -p -j 4 -r '(.*).(png)' cjxl.exe -e 6 -d 0.5 --faster_decoding=1 `{0`} `{1`}.jxl ``` nice and simple, it will encode the jxl file beside the png source file, I dont like using an encode directory since I often have files with duplicate names and I dont feel like making a script to create a new folder

fabian lookin post

nice and simple

jxl is doomed

fd has an --exec option which should execute commands in parallel

I do occasionally like exec, but lately I much prefer rust-parallel since its really powerful

one of the things you could do for instance would be (and I havent tested this so I may have messed up some ps stuff) ```ps fd --extension png | rust-parallel.exe -j 4 -r '(.*)\\(.*).(png)' mkdir encode\\`{1`} && cjxl.exe -e 6 -d 0.5 --faster_decoding=1 `{0`} encode\\`{1`}\\`{2`}.jxl ```

did you just answer your own question

then it wouldn't be parallel also can be done purely with fd `fd -e png -x cjxl "{}" "{.}.jxl" --num_threads 0` or with some options `fd -e png -j 8 -x cjxl -e 6 -d 0.5 "{}" "{.}.jxl" --num_threads 0` by default fd uses all threads, if needs less then `-j, --threads <num> Set number of threads to use for searching & executing (default: number of available CPU cores)`

sequential can be significantly slower so parallel should be preferred when possible

e9 is quite slow, especially for lossless, and for lossy it may be not better than e6/e7

I just use e4 for lossless myself

but, for strong lossless compression, yeah, e9 is better, but much slower

thanks again afed the command works like a charm

about 7% diff on average on random image sets and 2-4% on photo-like ones

nope, same quality, just one encoder per thread would be more efficient because cjxl is not fully parallelized (except for the fastest mode)

I remember posting a batch script ages ago when someone last asked, can't find it though

I love how overengineered everyone made their awnsers, including Lixus awnsers instead of a 1 line cmd command Would've dropped mine in the pile but not home right now

Wonder if a "High Threading" mode could be useful, disabling the singlethreaded options and lowering group size to increase multithreading

it will work if the encoder doesn't need anything outside the block, and this is true (or almost) for the fastest mode with streaming but, for slower modes not everything can be fully parallelized, i think even in theory (not just in current implementation)

not *everything* that the encoder does, no

but it certainly could be better than now

i.e. modular tree learning could be parallelized significantly more, but it's also a pain to write

33% space savings for free is a pretty huge deal...

I have read that the smallest possible JXL file is 12 bytes. Does anyone have such a file, as an example?

uhhh maybe <@794205442175402004> still has the one we crafted around?

`data:image/jxl;base64,/wr/BwiDBAwASyAY`

It was an adventure but a few months ago I dug up the old base64 encoding from Jon's Twitter, glad I kept it around

TY

It was made by using all the default values so most are 0 You can see an explanation of another JXL art file (With actual content) here https://discord.com/channels/794206087879852103/824000991891554375/901846536487579658

needs to be updated <:FeelsReadingMan:808827102278451241> https://github.com/mathiasbynens/small/pull/116

wow `/wp/`? All the small files I've looked at today started with `/wr` in base 64.

https://github.com/mathiasbynens/small/pull/125

I'd almost say the hand decoded file and smallest file should be pinned or placed in a 'resources' channel

I had a thought, couldn't individual layers of a layered JXL, or individual frames of animated, be loaded via a partial http request? Yet another use of partial requests alongside thumbnails and resized versions

I don't think we have a way to send the relevant offsets in advance for layers. For frames it can be done with a `jxli` box (frame index), but it indexes only keyframes so cannot be used for layers which are inherently not keyframes...

is there still any modes with smoother progression in current jxl, not just by blocks? i mean not browser implementation, but through djxl i was trying to compare with sqz, looks like it's better than progressive lossless jxl, but doesn't compete with normal lossy compression

like this sqz ? https://encode.su/threads/4183-SQZ-Low-complexity-scalable-lossless-and-lossy-image-compression-library

I was wondering why the lossless jxl lost a bit of the progressive flexibility of FLIF and FUIF ? what makes the new codec less progressivable ?

at the time, I liked this online demo very much : https://uprootlabs.github.io/poly-flif/

this might explain a little bit? https://discord.com/channels/794206087879852103/794206170445119489/1093762728562270219

interesting, but your question was for animations ? in the FLIF polyfill showcased in the link above, the result obtained for truncated still images (where only 10% is loaded) is stunning

I remember hearing somewhere that the "Squeeze" transform in lossless Modular JXL is not able to be progressively decoded that well or something like that. **EDIT**: I could be wrong as I'm looking though messages with "Squeeze" in them and seeing what looks like people saying the reverse?

yeah, i mean this codec jxl has a smooth progression like in the video below, but I'm not sure if it's still there in the current implementation i think it's because it's not optimal for browsers https://youtu.be/wN2Es9_hz7M

smooth progression for varDCT but not so smooth for modular (lossless)

seems like Squeeze is what makes progressive possible in modular mode however it's really inefficient at progressive lossless and ends up with large file sizes

ok, I'll ask <@794205442175402004> again 😄 only him, the creator of all losslesses could answer us !

for vardct also I haven't found how it works, even with all enabled options for progression, it's still by blocks

found this : https://discord.com/channels/794206087879852103/848189884614705192/1158994606797619250

yep, varDCT is just a generalization of block coding (and some refinements to limit block edges)

even with `--allow_partial_files` also it's not possible for all sizes, the first preview is only available after a certain percentage of the file `Input file is truncated and there is no preview available yet.` unlike sqz which has byte-by-byte granularity (maybe except for the header at the beginning)

some examples https://discord.com/channels/794206087879852103/803645746661425173/1193142742394613822

It depends a lot on the kind of image: for photographic, the gap shouldn't be that huge, for nonphoto, the gap is often large because the main coding tools for nonphoto (palette, lz77, patches, fancy MA trees) cannot be combined (at least not easily and not in the current libjxl encoder) with Squeeze.

LQIP was the name, removed from libjxl but JXL Oxide can do similar Lossless is still block by block https://discord.com/channels/794206087879852103/1065165415598272582/1167115137442918450 But lossy has the progressive scan, same as the video you linked, before filling in final blocks https://discord.com/channels/794206087879852103/1065165415598272582/1176210136612745347

yeah but lqip is not that, lqip is a small preview before the full image is loaded but smooth rendering while decoding the whole image is most likely what doesn't work

LQIP is a somewhat ambiguous/vague term, technically a LQIP could be done in a different form that's more closer to the performance oriented/focused progressive decoding that libjxl currently does, here's the middle of a discussion from a few years ago relating to this https://discord.com/channels/794206087879852103/794206170445119489/887306337048989706

I do think it would be nice if libjxl got full support for encoding and decoding the type of LQIP that was discussed a few years back where it's just a step before the 1:8 "MQIP" image that is decoded currently as it would mean JPEG XL could act as a full replacement for things like BlurHash or ThumbHash

would probably work really well as a optional thing you could pass into libjxl when encoding to add a LQIP (1:64 or something similarly small) so websites and companies could just add the flag to their encoding CDN and replace things like BlurHash or ThumbHash without affecting websites that don't like the distracting nature of really low res LQIPs

i believe lqip should work with vardc with some options?

I remember seeing progressive passes being messed around with in past messages in this server but I never really looked more into it

does anyone know if in libjxl does the current encoder support adding in such a low quality pass? and does the current decoder accept it during progressive decoding?

there are some options, but I haven't tested for lqip either ``` -p, --progressive Enable (more) progressive/responsive decoding. --progressive_ac Use the progressive mode for AC. --qprogressive_ac Use the progressive mode for AC with shift quantization. --progressive_dc=num_dc_frames Progressive-DC setting. Valid values are: -1, 0, 1, 2.```

`--progressive_dc 1` adds squeezed modular LF frame

this should be enough for low quality preview of partial image

also a new j2k encoder release (htj2k), but no quality improvements, just optimizations sqz is basically what a modern j2k-like codec should be, with similar quality scaling https://www.phoronix.com/news/OpenJPH-0.10-Released

for non-photo sqz is not that good, too blurry, even jpeg can keep more textures and details (except at very low bitrates)

jpeg, jxl, sqz

and progressive truncated jxl lossless with the same size <:kekw:808717074305122316>

Good for me autism

See how the bitrate got decressed

https://vm.tiktok.com/ZGe2cgBj3/

Interestingly, this JXL file decodes fine in jxl-oxide and also in jxlatte, but djxl aborts if you run it with --disable_output

I wonder if it's a djxl bug

libjxl-based decoders don't have any issues

Giving it an output format decodes fine, so yeah, just a djxl bug

How do I tell cjxl what colorspace a `.pfm` file is in?

it's in Linear sRGB, but it's assuming nonlinear sRGB

do we assume nonlinear by default for pfm? not sure we should

perhaps but I'm really wondering how to specify the space for untagged input

the only color options I see in cjxl -h -v -v -v is `-x colorspace=foo` which iirc sets the tagged space for XYB images

rather than specifying the input space

it does specify the input space

in my case I have a `.pfm` image that I'd like to (lossily, XYB, etc.) encode to a jxl file. The input is Linear sRGB, but I want the actual JXL file to be tagged as nonlinear sRGB.

but only as a “hint” in case no colorspace can be inferred from the file, so it’s pretty much incapable of overriding it for PNGs

the tagged space is somewhat unimportant because it's an XYB encoded file

(or maybe for completely untagged PNGs but I wouldn’t bank on it)

Is there any way to tell cjxl that "the input is linear, but I want it tagged as sRGB" for an XYB image?

not that I know of

theoretically it sounds possible, as the input image space doesnt' have to match the tagged space if `uses_original_profile == false`

I wonder if I could theoretically convert the .pfm to lossless jxl with `-x color_space` to set it to linear

so I now have a lossless JXL file that has color tags, with identical pixel data to the pfm

and then from there run `cjxl`

i.e.

``` cjxl -d 0 -e 1 -x color_space=RGB_D65_SRG_Rel_Lin input.pfm intermediate.jxl cjxl -d 1 -e 7 -x color_space=RGB_D65_SRG_Rel_SRG intermediate.jxl output.jxl ```

I suspect that the second `-x color_space` will do nothing

because the input is already tagged so there is no need for a hint

if that's the case, it shouldn't be called "dec hints"

since the documentation in `cjxl -h -v -v -v` specifies that it tags the output file. i.e. hint to the decoder

it’s a hint to the decoder of the input

well, then it's very misleading

since it's not at all a hint, it's a prescription

in what way is it a prescription rather than a hint?

because it's declaring the input space to be blah

only if the input isn’t already tagged

a "hint" implies that it's a suggestion that may optionally be ignored

but for lossy VarDCT/XYB it cannot be ignored, it is critical information

But in either case, is there any way to tell cjxl how to tag the output space

for XYB?

without actually converting it *to* that space with a CMS engine and then passing that data to cjxl

I don’t know of such a way

it could indeed be useful

Does anyone know of any interesting non-libjxl jxl encoders? I think there's one called hydrium, but I'm not sure if there are any others.

as far as I'm aware Hydrium is the only one, and it's still super WIP

(disclaimer: I am the author of hydrium)

looks like it does VarDCT? Does it do any Modular beyond what is needed for the DC?

It's currently locked to VarDCT, and has no quality setting

currently targets ssimulacra2=~85 for photographic content

neato

the primary design goal is that it's a streaming encoder

and has ultra-low-memory usage

I want to make a modular mode encoder, to try stuff libjxl doesn't do.

yea, the memory usage was the primary design goal

I'll probably accidentally write something that uses a gigabyte per megapixel

apart from hydrium, there's also https://github.com/etemesi254/zune-image/tree/dev/crates/zune-jpegxl - this one is losless modular mode encoder

aha

I mean, technically, what's now `-e 1` used to be a standalone encoder xD

(and can still be used that way)

I'm still somewhat convinced it should be possible to do much better for modular encoding, both in terms of speed-for-compression and in absolute terms

but who has the time...

Not far off libjxl at e9 then ;P

lol

it's ignoring it for PFM input too though, for some reason

is that a bug?

arguably

Update: the bug was me typing `-x colorspace=` and not `-x color_space=`

oy, maybe we should try to detect this sort of thing somehow

like keeping track of accesses to hints and printing “the following hints were not used:”

it does print a weird warning and only if the debug is enabled

(although, in this specific instance, it wouldn’t make it immediately clear why it was unused)

``` $ cjxl -x colorspace=RGB_D65_SRG_Rel_Lin input.pfm test.jxl JPEG XL encoder v0.10.0 875e72dd [AVX2] ./lib/extras/dec/color_hints.cc:61: Ignoring colorspace hint Encoding [VarDCT, d1.000, effort: 7] Compressed to 750.6 kB (1.362 bpp). 2100 x 2100, 5.565 MP/s [5.57, 5.57], 1 reps, 8 threads. ```

Likewise `cjxl -x dsgjkfgj=RGB_D65_SRG_Rel_Lin` returns `./lib/extras/dec/color_hints.cc:61: Ignoring dsgjkfgj hint`

oh, it should at least say “unrecognised”

`Ignoring unrecognised hint “colorspace”` would be much more helpful, wouldn’t it

IMO this should: (1) say something like `Ignoring unrecognized hint: "colorspace"` (2) possibly hard-fail

if it's not going to hard-fail it shouldn't silently ignore it without printing the warning on Release mode

as this is a debug-build-only warning

for context, I just encoded a 2100x2100 image to jxl with hydrium and the libhydrium memory usage itself capped out at 1,338,213 bytes

aka less than 1.5 megabytes for the encoding process

the actual JXL file is 2M, for context

and the PNG input is 15M

nice

You can use one-frame mode, and it gets much hungrier

the same image takes 61 M of ram to encode in one-frame mode

me: about to bring up an example that I just used it for to compare

also me: wow that's a segfault also me: okay that's fixed also me: and it's distorted

I said WIP 😅

oop

but ye all the bugs I just encountered have been fixed <:poggies:853085814032302122>

hooray

<@794205442175402004>

I managed a 2,96mbps jpg

FB algorithm

Is the worst shoot of one I did

Normal gs camera

O well shoot

So a deblur 0,65

<#806898911091753051>

Took a manga series, original format was .cbz containing .png, ~200 pages a volume, 61 volumes total, 1,500x2,250 black & white pictures w/ the cover being in color. Total size of 12.6 GiB I used this bash command to convert them into distance 1 effort 9 .jxl in an lzma2 .cb7 : ``` for i in *.cbz; do 7z e -o"${i%.*}" "$i"; rm "$i"; for j in "${i%.*}"/*.png; do ffmpeg -i "$j" -c libjxl -effort 9 -distance 1.0 "${j%.*}.jxl"; rm "$j"; done; 7z a -t7z -mmt=1 -mx=9 -ms=on "${i%.*}.cb7" "${i%.*}"; rm -r "${i%.*}"; done ``` The result was 7.0 GiB, ~100 MiB of which was lzma2/7z. So a total of 5.5 GiB saved with jxl

oh and for those that don't know, .cb* is a comic book * archive. literally just a .zip or whatnot with images sorted by filename

So easy to mess with I love it

It did take forever tho on my poor i5-7300U

Look like you got some high quality source, otherwise most manga are JPG isn't it. How long is forever? 3 days?

i dont remember how to convert xyb jpeg to png/ppm for example, but for normal viewing without icc profiles support?

At a guess, I'd assume they used lossy *because* the source files were already compressed. If it were truly monochrome and crisp, I'd expect lossless could've been competitive with lossy

I haven’t tried it but ImageMagick should work, something like: ```console $ convert xyb.jpeg -profile .../sRGB.icc regular.png ``` where one sRGB profile you can use, if you have colord installed, is `/usr/share/color/icc/colord/sRGB.icc`

argyll also has one somewhere if you have it

<@532010383041363969> I've beaten your codec

https://www.reddit.com/r/DIY/s/fxoeRvR1ve

We still have no idea what you mean

Whatsapp audio encoder

https://discord.com/channels/1068042250145251390/1068068842938568724

You're probably right about that, fortunately the quality is good enough to where I can't really see any artifacts

<@532010383041363969> you can't beat this

Even with 1000 core

there's a few ways but the easiest way to do it is: ``` ffmpeg -i input.jpg -vf libplacebo=color_trc=13:color_primaries=bt709:format=gbrp output.png ```

you can always `djpegli input.jpg output.png` but that just attaches the XYB color profile rather than actually converting the colorspace

but you can then feed that to the ffmpeg command above instead

Beat pizza with CPU cores? What?

It's a secret fab technology, makes pizza to be forever hot, lol

yeah, that's what I did `djpegli input.jpg output.png`, but that requires support for color profiles, which I need to avoid

<@1034873369314730065> facts this file contains the old values of Butteraugli used in libjxl 0.8.1 by Jon Sneyers

Is a reverse engineering

Sdzabaka is OK for me MERGE

Jon i can't access tik Tok anymore

Veluca do you like what you seeing?

I'll compress what you want send me image

Sento il tuo respiro, mi scorri dentro, Come fiume calmo, profondo sentimento. Nel cuore intrecciato, come un dolce canto, La tua presenza, un eterno incanto. Riflessi negli occhi, come stelle brillanti, Nel buio della notte, sempre costanti. Mi avvolgi con abbracci, come onde in mare, Il tuo calore, un rifugio da amare. Sei il suono dolce in un silenzio intenso, Il segreto sussurro, il nostro incanto immenso. Attraverso le stagioni, come foglie al vento, Mi scorri dentro, in ogni momento. E nell'abisso dell'anima, tu sei il colore, La penna che scrive la nostra storia d'amore. Mi scorri dentro, come poesia viva, In un viaggio eterno, l'anima captiva.

I don't know what I'm doing

Jon there is the Notizia

Thanks for allowing me

Here we ARE

Full HD version

We have full HD Whatsapp now

HD vS standard Definition

HD vs Standard Definition

Jon please tell me if what I'm doing is right

I managed to improve color

Test foto Normal

spidermario

Ok

See if is an improvement spider

Jon i fatigued

I did finished the JPG end bitstreams

Try to further optimize

Optimizing the JPG codec involves various considerations, typically related to balancing compression efficiency and image quality. Here are a few general strategies: 1. **Quantization Tables:** Adjusting the quantization tables can impact file size and image quality. Experiment with different settings to find the right balance for your needs. 2. **Huffman Coding:** Fine-tuning Huffman coding parameters can influence compression efficiency. Some tools allow you to adjust Huffman tables for optimization. 3. **Chroma Subsampling:** Adjusting chroma subsampling ratios can reduce file size, but be cautious as it may affect color accuracy. Common subsampling ratios are 4:4:4, 4:2:2, and 4:2:0. 4. **Entropy Coding:** Optimize entropy coding parameters, considering the trade-off between compression ratio and computational complexity. 5. **Baseline vs. Progressive:** The baseline format loads from top to bottom, while progressive loads in multiple passes. Choose the appropriate format based on your application. Remember, optimizing the JPG codec requires a good understanding of the trade-offs between compression and image quality. It's advisable to use specialized tools or libraries that provide fine-grained control over codec parameters.

Optimizing a JPG codec involves enhancing its efficiency in terms of compression and decompression. Here are some general strategies: 1. **Quantization Tables:** Fine-tune quantization tables to balance image quality and file size. Adjusting these tables can significantly impact compression efficiency. 2. **Huffman Coding:** Optimize Huffman coding for entropy encoding. Efficient Huffman coding can contribute to better compression ratios. 3. **Chroma Subsampling:** Experiment with chroma subsampling ratios. This can reduce file size, but be mindful of potential color artifacts. 4. **Progressive JPEG:** Implementing progressive encoding can provide a better user experience, allowing a gradual rendering of images. 5. **Lossless Compression:** Consider supporting lossless compression for cases where preserving every detail is critical. Optimizing a codec requires a deep understanding of the JPEG standard and its trade-offs. Depending on your specific goals, you might need to experiment with different parameters to find the right balance between file size and image quality.

Huge upgrade

This isn't related to JXL, though... If I'm understanding you're talking about normal Jpeg

someone should make a bmp wikipedia page for jxl

New version out

<@204686867499581440> please help https://crowdsource.google.com/csf/?label_str=https://photos.app.goo.gl/ofMpFrrnKfKNY3TEA&lang=it_IT&utm_source=google-photos&utm_campaign=empty-search-results#/contribute/image-label-verification/categories.

Don't do autobus and nuvola that I did it now CLOUDS

Pashi your problem was solved by 65%

Re encoding

Bpo 0.036

Jon wait please one minute

I think even more Discord is generating the file for you

I did continue

Internet is slow

Jon is encoding

Jon ese es el link https://chat.openai.com/share/e82e0be1-c47c-4a4b-823a-f73622581f2c

Jon please do 60 better

I'll help Jon with a Font to program

Most compressed video ever

Same bpp but from Meta reencoded mu-ltiple times

It seems like a script for batch processing comic book archive files. Let's break it down: 1. `for i in *.cbz`: Loop through all files with the `.cbz` extension in the current directory. 2. `7z e -o"${i%.*}" "$i"`: Extract the contents of each `.cbz` file into a directory with the same name (excluding the extension). 3. `rm "$i"`: Remove the original `.cbz` file after extraction. 4. `for j in "${i%.*}"/*.png`: Loop through all PNG files in the extracted directory. 5. `ffmpeg -i "$j" -c libjxl -effort 9 "${j%.*}.jxl"`: Convert each PNG file to the JPEG XL format using FFmpeg. 6. `rm "$j"`: Remove the original PNG file after conversion. 7. `7z a -t7z -mx=9 -ms=on "${i%.*}.cb7" "${i%.*}"`: Create a new 7z archive with the same name as the original `.cbz` file, containing the converted images. 8. `rm -r "${i%.*}"`: Remove the extracted directory and its contents. This script essentially converts images within comic book archives from PNG to JPEG XL and repackages them into a new 7z archive while cleaning up temporary files. If you have any specific questions or if you need assistance with something related to this script, feel free to ask!

Jon another bitstream

https://youtu.be/wO4tAJpmM0U?si=fZtb-zApRgAxbycx

Other bitstreams

Other bitstreams I genenerated jXl for you

Recent Bitstreams

Latest opus encoder

I'm still working

Tell me what to do

New chat Jon Sneyers

https://chat.openai.com/share/1ee085d3-0f75-434e-830e-9f30506d78fc

CBZ to JXL Conversion

That's my take

Tik tok Full HD quality

They are re encoded everything

Fab multipass technology

<@456226577798135808> Jon I've extended the link, please visit

Please devs download until is taken

https://chat.openai.com/share/a3029529-df2e-433d-951c-e88aa1deee56

Jon here s newer fonts for you

Jon, new font for you

Jon a typeface created with artificial intelligencs

I published complete version

Thanks to my neIghbour we now have improved JPG conpression

<@532010383041363969>

Meta servers got blocked

Please jyrki help

spam

https://chat.openai.com/share/338f5733-18ba-413b-a12d-c9ce0ffcd64d

<@794205442175402004>

fab could you maybe move your stuff to <#840831132009365514> and/or <#806898911091753051> ? And maybe stop tagging people with things that are not a question nor something comprehensible.

Better not explicitly mention stuff on the public web that might wake sleeping dogs. But of course you can always invite developers to the jxl discord 🙂

https://small.fileditchnew.ch/s2/XCgQmmUCJnsRwRgBPlBi.zip

Fully working phone.zip

Three resolutions me

<:retiring:717702574676901908>

is a joke

How can mute myself for a month

https://github.com/fabiorug/input-ai-font-paid

wb last update

19:47

This is still unrelated

<:NotLikeThis:805132742819053610>

This are all new fonts 8 pack

https://small.fileditchnew.ch/s2/eFpbRVAHaVwNVJmbHmXZ.zip

Attualmente, non ho informazioni specifiche sulla versione premium di ChatGPT-4 o sui plugin disponibili. Tuttavia, posso darti alcune considerazioni generali. Se la versione premium offre accesso a plugin, compresi quelli basati su modelli come DALL-E per la generazione di immagini, potrebbe valere la pena esaminare l'upgrade se sei interessato a sperimentare con la generazione di immagini insieme alle funzionalità di chat. Quando si tratta di generazione di immagini, modelli come DALL-E sono noti per la loro capacità creativa. Tuttavia, la scelta tra l'opzione premium di ChatGPT-4 e MidJourney dipenderà dalle tue esigenze specifiche e dalle tue preferenze lavorative. Potresti considerare i seguenti aspetti: 1. **Usabilità:** Valuta la facilità d'uso di entrambe le opzioni e come si integrano nel tuo flusso di lavoro. 2. **Personalizzazione:** Controlla le opzioni di personalizzazione offerte da ciascuna soluzione per adattarle alle tue esigenze. 3. **Qualità delle Immagini:** Se hai esperienze passate positive con MidJourney, potrebbe essere un punto a favore. 4. **Feedback degli Utenti:** Cerca recensioni o feedback da altri utenti che hanno sperimentato sia la versione premium di ChatGPT-4 che MidJourney. 5. **Aggiornamenti Futuri:** Considera anche la possibilità di aggiornamenti futuri e nuove funzionalità che potrebbero essere rilasciate. Ricorda che le tecnologie e le offerte possono evolversi rapidamente, quindi potrebbe essere utile verificare le informazioni più recenti o contattare direttamente l'assistenza clienti di OpenAI per dettagli aggiornati sulla versione premium e sui plugin disponibili

Currently, I have no specific information about the premium version of ChatGPT-4 or the available plugins. However, I can give you some general considerations. If the premium version offers access to plugins, including template-based ones like DALL-E for image generation, it might be worth looking into upgrading if you're interested in experimenting with image generation alongside chat features. When it comes to image generation, models like DALL-E are known for their creative ability. However, choosing between the premium option of ChatGPT-4 and MidJourney will depend on your specific needs and work preferences. You might consider the following aspects: 1. **Usability:** Evaluate the ease of use of both options and how they integrate into your workflow. 2. **Customization:** Check the customization options offered by each solution to fit your needs. 3. **Image Quality:** If you have positive past experiences with MidJourney, this could be a plus. 4. **User Feedback:** Look for reviews or feedback from other users who have experienced both the premium version of ChatGPT-4 and MidJourney. 5. **Future Updates:** Also consider the possibility of future updates and new features that may be released. Please remember that technologies and offerings can evolve rapidly, so it may be helpful to check the latest information or contact OpenAI customer support directly for up-to-date details on the premium version and available plugins

In the serene tranquility of a sunlit day, the leaves, delicately caressed by the whispering wind, perform a graceful ballet among the majestic branches of the trees. The azure sky, a vast canvas, stretches infinitely overhead, adorned with cotton-like clouds drifting lazily by. The warm embrace of the sun, akin to a gentle lover's kiss, bathes the earth in its golden glow, casting a mosaic of shadows beneath the foliage. Nature itself becomes a grand symphony, with the rustling leaves and the occasional chirping of birds forming harmonious notes in the air. Each tree stands tall, a silent guardian witnessing the timeless dance of seasons. The scent of earth and blooming flowers weaves through the air, creating a fragrant tapestry that lingers as a testament to the beauty of the moment. As the day unfolds, the sun's journey paints a vivid mural across the sky, transitioning from the soft hues of morning to the vibrant brilliance of midday and eventually casting a warm, golden-orange hue during the serene hours of evening. The landscape becomes a living, breathing canvas, capturing the essence of a fleeting yet eternally recurring masterpiece. Within this enchanting setting, a profound sense of serenity envelops the surroundings. It is as if time itself slows down, inviting individuals to immerse themselves in the gentle rhythm of the natural world. Every rustle of a leaf, every subtle shift of light, becomes a part of a larger, intricate choreography that unfolds without haste. This idyllic scene becomes a sanctuary of peace, a refuge where one can escape the complexities of life and find solace in the simplicity of nature's embrace. In this moment, every breath is a celebration, and every rustle in the wind is a whisper of timeless wisdom. It is a small ritual of peace, where the beauty of the world invites contemplation and a connection to something far greater than ourselves.

the dedication I give to improve Codecs

Warning the video is a bit autistic

J

https://www.facebook.com/100064789743870/posts/pfbid0j71Q8Mb8vnqsuV67CgETiSCxVnkb34hzmyFh1ek24ue6cBmxPfKvRMQBUsFyovDWl/

<@794205442175402004> here's the person who sacrificed for you

🕡💌📧🕑

Muting you for a day, fab. You're producing too many incomprehensible/irrelevant messages. I warned you yesterday, but you ignored that warning.

.

<@288069412857315328> Is there a way to start a new spline on this page? Or is it kept simple and only 1 continuous spline https://grass.moe/catmull_rom/

it's just one

re all , converted losslessly jpg to jxl - Irfanview viewer shows increase in number of colors is it normal?

yes, JXL decodes the same information differently

The jpg file it converts back into should be identical though

this is a quirk of JPEG itself. JPEG decoding is not fully specified, i.e. there's more than one correct way to decode a JPEG file

when you transcode JPEG to JXL, and then decode the JXL, you'll always get **a** correct decoding of that JPEG

but you won't necessarily get the **same** correct decoding that another correct JPEG decoder will give you, such as libjpeg-turbo

(you’ll get a better one, arguably)

I did compare the 'standard' jpeg decode to a JXL trannscode in Krita a week ago. The largest differences were around the highest areas of noise, but it seemed to mostly shift color shades and edges of color banding by a few pixels, so probably just decoder differences rather than JXL improvement in that case

This has a curious side effect. If I open a transcoded jpeg-jxl in IrfanView, downscale it and save it as a jpeg with a particular quality setting, it is usually slightly smaller than if I load the original jpeg file rather than the transcoded one.

makes sense

Ecco una risposta che ho generato usando il nuovo Bing, il motore di risposte basato su intelligenza artificiale. Fai clic per visualizzare la risposta completa e provalo anche tu. https://sl.bing.net/iX11vigMaiG

12,76C

Is fab an AI?

almost, he is an Italian

lol

<@416586441058025472> This is my final warning. Off-topic or incomprehensible/nonsensical messages are annoying and are perceived by many including me as spammy. Stop it or I'll mute you again, for a week this time.

?

<https://github.com/libjxl/libjxl/issues/2543>

Ok 🎐

I review 1000 issues

Done

Best developers in Github https://github.com/kaspernordkvist/uncut_plan8

I resumed to

Scientific link https://www.meteogiornale.it/2024/01/peggiore-tormenta-neve-verso-italia-conseguenze/amp/

Forse intendevi meteo Giarre, un comune italiano della città metropolitana di Catania, in Sicilia. Il meteo per domani a Giarre è il seguente¹²: - Cielo **parzialmente soleggiato** con temperature minime di **7°C** e massime di **15°C**. - Possibilità di **14%** di precipitazioni e un avviso di **fenomeni meteorologici di moderata intensità** fino alle **1:59** di domani¹. - Vento da **SO** con intensità di **6-9 km/h**. - Sorgere del sole alle **7:13** e tramonto alle **17:04**¹. Per maggiori informazioni, puoi visitare i siti [ILMETEO.it](^1^) o [METEO.IT](^2^). 😊..

Le previsioni del tempo per la prossima settimana a Giarre sono le seguenti¹²: | Giorno | Tempo | Temperatura minima | Temperatura massima | Precipitazioni | | --- | --- | --- | --- | --- | | Sabato 20 | Parzialmente nuvoloso | 7°C | 15°C | 14% | | Domenica 21 | Sereno | 5°C | 14°C | 0% | | Lunedì 22 | Poco nuvoloso | 7°C | 14°C | 10% | | Martedì 23 | Molto nuvoloso | 10°C | 14°C | 40% | | Mercoledì 24 | Coperto | 10°C | 16°C | 80% | | Giovedì 25 | Variabile | 13°C | 18°C | 60% | | Venerdì 26 | Parzialmente nuvoloso | 13°C | 19°C | 20% | Per maggiori dettagli, puoi visitare i siti [ILMETEO.it](^1^) o [METEO.IT](^2^). 😊. Origine: conversazione con Bing, 15/1/2024 (1) Meteo GIARRE Previsioni fino a 15 giorni » ILMETEO.it. https://www.ilmeteo.it/meteo/Giarre. (2) Previsioni Meteo Giarre Oggi - Fino a 15 Giorni 3B Meteo. https://www.3bmeteo.com/meteo/giarre. (3) Meteo GIARRE Previsioni fino a 15 giorni » ILMETEO.it. https://www.ilmeteo.it/meteo/Giarre. (4) Previsioni Meteo Giarre Oggi - Fino a 15 Giorni 3B Meteo. https://www.3bmeteo.com/meteo/giarre. (5) Previsioni meteo Giarre tra 7 giorni | METEO.IT. https://www.meteo.it/meteo/giarre-7-giorni-87017. (6) Meteo Giarre - meteoblue. https://www.meteoblue.com/it/tempo/settimana/giarre_italia_2524618. (7) Previsioni meteo Giarre - Fino a 15 giorni | METEO.IT. https://www.meteo.it/meteo/giarre-87017.

Il meteo per sabato a Giarre, Sicilia, è il seguente¹²: - Cielo **parzialmente nuvoloso** con temperature minime di **7°C** e massime di **15°C**. - Possibilità di **14%** di precipitazioni e un avviso di **fenomeni meteorologici di moderata intensità** fino alle **1:59** di domani¹. - Vento da **SO** con intensità di **6-9 km/h**. - Sorgere del sole alle **7:13** e tramonto alle **17:04**¹. Per maggiori informazioni, puoi visitare i siti [ILMETEO.it](^1^) o [METEO.IT](^2^). 😊. Origine: conversazione con Bing, 15/1/2024 (1) Meteo GIARRE Previsioni fino a 15 giorni » ILMETEO.it. https://www.ilmeteo.it/meteo/Giarre. (2) Previsioni Meteo Giarre Oggi - Fino a 15 Giorni 3B Meteo. https://www.3bmeteo.com/meteo/giarre. (3) Meteo GIARRE Previsioni fino a 15 giorni » ILMETEO.it. https://www.ilmeteo.it/meteo/Giarre. (4) Previsioni Meteo Giarre Oggi - Fino a 15 Giorni 3B Meteo. https://www.3bmeteo.com/meteo/giarre. (5) Previsioni meteo Giarre - Fino a 15 giorni | METEO.IT. https://www.meteo.it/meteo/giarre-0087017. (6) Previsioni meteo Giarre - Fino a 15 giorni | METEO.IT. https://www.meteo.it/meteo/giarre-87017. (7) Previsioni meteo Giarre - Fino a 15 giorni | METEO.IT. https://www.meteo.it/meteo/giarre-prossimi-giorni-87017.

Are you ok?

Re yes

Qualità dell'aria 69

There is no escape

you don't seem to be "ok". could you explain in english without using ai the purpose/reasons for recent messages you have been posting such as https://discord.com/channels/794206087879852103/794206170445119489/1196503656653131867 and https://discord.com/channels/794206087879852103/794206170445119489/1196514990316859514 and more ?

He said he has autistic disorder, so probably it is impossibe to understand

Ciao, questo è Bing. Mi piacerebbe giocare a tris con te, ma non ho questa funzionalità. 😔 Però, posso offrirti alcune alternative. Ho trovato alcuni siti web dove puoi giocare a tris online gratuitamente contro un computer o altri giocatori. Ecco alcuni link che potrebbero interessarti: - [Tris ️ Gioca su CrazyGames](^1^): Un sito dove puoi giocare a tris su una griglia di dimensioni diverse, con diverse modalità di gioco e livelli di difficoltà. - [Gioca a Tris Online Gratuitamente Contro un Computer o Multigiocatore](^2^): Un articolo che spiega come giocare a tris, le regole, le strategie e i vantaggi di giocare online. - [Tris: regole e tattiche per vincere](^3^): Un altro articolo che illustra le regole e le tattiche per vincere a tris, con alcuni esempi e consigli. Spero che questi siti ti siano utili e ti divertano. Buon divertimento! 😊. Origine: conversazione con Bing, 15/1/2024 (1) Tris ️ Gioca su CrazyGames. https://it.crazygames.com/gioco/tic-tac-toe. (2) Gioca a Tris Online Gratuitamente Contro un Computer o Multigiocatore .... https://playordown.com/it/online-games/giocare-a-tris/. (3) Tris: regole e tattiche per vincere - Giochi STARS. https://giochistars.it/news/tris-regole-e-tattiche-per-vincere/. (4) Tris - Gioca Gratis Online su SolitaireParadise.com. https://www.solitaireparadise.com/it/gioco/tris.html. (5) Getty Images. https://www.gettyimages.com/detail/photo/tick-tack-toe-on-chalkboard-royalty-free-image/1134456281.

<@&807636211489177661> these are spams, but even more than that: scams Didn't know if I had to ping admins but I feel these messages need to be deleted

fab wrote a lot of messages lately but never scams, perhaps he got hacked

it looks like fab is just copy and pasting from AI chatbot chats or something like that

I don't think so, he is just using these chats as personal pastebin

his computer's been hacked several times before, so it's possible

from the twitch chat

<@794205442175402004> fab is doing the thing

please help

I'm not thething

I can't mute him, only kick him and I don't want to do that

Mi chiamo charles

I don't know if the *Silenced* role doesn't anything

Non ho 2FA al Momento

users can block another, it hides his messages for as long as block is in place and indiv messages can still be displayed, just rescind later

yea, I don't like blocking people though

cause it just hides the messages, doesn't remove them

I find the block feature mostly useful to prevent receiving unsoclited DMs

blocking cant get really buggy because fab will sometimes post up to 80 messages and it will bug out clients ~~I wonder how I know~~

FABIAN.

PLEASE STOP IT WITH THE SPAM

Muted fab for a week.

<@416586441058025472> I hope a week will be enough to resolve whatever the problem is.

he asked for a month why not mute for a month

I've read that jbrd is supposed to be compressed but JFIF marker is often clearly visible inside jbrd. Is the compression applied only if the size of jbrd reaches some threshold? As jbrd is used only for reconstruction, it wouldn't harm if it was always compressed.

I think it's always compressed into a Brotli stream, but it can choose to not compress small enough inputs.

This would make sense if Brotli was unable to compress the data so it would store the raw bytes instead. That's what file archivers do if they encounter a file where the compression doesn't work.

Compression is on by default but can be disabled for compatibility, IIRC

This can be done for other boxes but this setting does not apply to jbrd.

Ah, right

How do I get lossless jxl files in the smallest file size possible? Is it better than Free Lossless Image Formats (FLIFs) files?

Are you the same user as on Reddit by any chance?

oh

It's fine either way, just feel like it's very similarly worded haha

you got an answer?

Jon made FLIF, and JpegXL is the successor so files *should* be smaller Using the `cjxl` tool from Github, generally `-d 0 -e 9 -g 3 -E 3 -I 100` will get you 99% of the way ||For the smallest size currently possible, you could try `-d 0 --allow_expert_options -e 10` BUT it's extremely intensive on both CPU and RAM, so can't encode anything over a few hundred pixels in both dimensions without crashing||

I only saw the reddit once when it was first posted, my memory isn't that great xP

does the cjxl for making image sequences of lossless jxl from videos? can you send me the link?

I wouldn't reccomend lossless from videos since they're already compressed

Well then, how do I make image sequences of lossless jxl files from videos?

entire videos

extract every single frame as lossless jxl

if its mjpeg footage (ie from cctv) you could losslessly convert all images into lower sized frames smaller than than the corresponding jpegs

I have H.264 amd H.265 videos

what's the use case

unless mistaken, i vaguely recall someone mentioning interframe compression as an optimzation for video (meant for cctv-style almost static footage), but you may find proper video codecs preferable here, especially if source wasnt originally a lossless noise-free capture

I just need a lossless format smaller than png files

use case?

already lossy h264 will compress very poorly to lossless jxl, no ifs or buts

sounds like you should just use lossless jpeg

Then how do I get that with my videos?

very carefully

regular jpgs are lossy

So is regular video

<:ackshually:587475701209497610>

Yes

I can still tell the difference

jxl with a very low distance (sub 1) could be better than lossless jxl but its still gonna be lossy. at this point just keep the original video

use case!!!!!!

I think FFmpeg is able to do this however I don't know how to use FFmpeg.

I would've assumed you need lossless because a program only accepts it, but JXL is hardly widespread yet Could try ranging from `-d 0` to `-d 1` on a single frame to see at what point you notice the compression

What about YUV videos (4:2:0, 4:2:2, and 4:4:4)? With limited color range and BT.709 matrix coefficients? And again, I need an alternative lossless format that's smaller than png files.

<:linusretire:707375442226708542>

just stick with png

How do I make lossless webp files, then? My videos are YUV.

bro

yeah uhh refer to the arch wiki

how I do it is extract to a PNG sequence first, then transcode that. Use powershell on windows. 1. create new folders called img-seq 2. `ffmpeg -i video.mp4 -c:v png img-seq\frame-%04d.png` 3. Use a parallel command runner, in my case I use rust-parallel on windows, you can invoke it like this ` ```ps ls -name | rust-parallel.exe -p -j 4 -r '(.*).(png)' cjxl.exe -e 4 -d 0 `{0`} `{1`}.jxl ```

I extract to a PNG first then use cjxl in parallel since it is much faster then just using ffmpeg

however you can just do `ffmpeg -i video.mp4 -c:v libjxl -distance 0 -effort 4 img-seq\frame-%04d.jxl`

I use it as a mezzanine

I detail it more here https://quackdoc.github.io/Post/jxlkindacool.html

I use 0.5 though with faster_decoding

I can't seem to figure out what -g -E and -I mean, do you have any explanations?

g is the group size, how many pixels it works on at once. Sometimes smaller can be better so it can better focus on that area of the image, but most times the biggest is best E lets the encoder compare R G B at once, instead of only looking at red to compress the red channel, ect I is how many of the pixels in g it looks at to make it's decisions All of them increase the encoding time and memory usage, or can be worse than other values, hence why they're not the defaults

ahhh makes sense

Thanks!

Some bonus info g 0 is 128x128, g 1 is 256x256, g 2 is 512x512 and g 3 is 1024x1024 I is a percentage, so 0 disables the learning entirely and speeds up encoding but lowers quality/increases size

Which one of these other lossless formats is the smallest? https://docs.google.com/spreadsheets/d/1ju4q1WkaXT7WoxZINmQpf4ElgMD2VMlqeDN2DuZ6yJ8/edit#gid=174429822

the one with the smallest number

I've discovered an interesting behavior if predictors are turned off. For some images, disabled predictors cause a massive drop in speed. That means the encoding speed is several times slower. That's kinda contradictory as I'd expect skipping predictors would speed everything up.

without looking at a profile I just have to guess, but I assume it makes tree learning much much slower

I need to do more testing but it seems that smooth color gradients trigger the issue.

spoiler : jxl is better than webp, png and avif lossless almost everywhere except for some problematic files (few of them, especially palettized) where webp takes the lead

BMF and EMMA does better but are not very convient formats

interesting... and not tested enough in the spreadsheet indeed

I am assuming it can be fixed 😄

I think I know what's up

fixed ? easy 🙂

haha

😛

Do I even have to say it

They have larger file sizes?

<https://github.com/libjxl/libjxl/pull/3151>

So they are smaller than jxl?

Well then, how do I make them?

I believe they are both windows-only binaries? might be wrong though

What about ppm images? Are they lossless?

ppm is lossless because they're actually just uncompressed

So it's bigger. Never mind

I can make a lossy ppm encoder 😝

Yea but that's not a property of the format

that's just chosing to throw out data

Anything is lossy if you don't care too much about the results ~~truncating~~

What about these? https://encode.su/threads/2459-EMMA-Context-Mixing-Compressor http://compression.ru/ds/

wouldn't <#805176455658733570> be more fitting?

<:Poggers:805392625934663710> https://github.com/libjxl/libjxl/pull/3155

but without streaming png support it's unfortunately will be less usable by average people

it still loads the image into memory, but it will use less memory to encode

I suppose we could in principle make the png loader streaming too, at least in the case where it's not Adam7.

Does it load the full image in a float32 buffer or only in a uint8/16 RGB(A) buffer and the conversion to planar float is already done in dcgroup sized tiles?

would be great, like libhydrium with libspng

I am not sure

but even just doing u8/u16 full image buffers and then tiling the float conversion would go a long way

If I recall the u8 to float was the main cause of memory usage for extremely large images, namely jpegs that are being lossily compressed