>>102105985
Left is better?

>>102106031
Left is better

right is more sovl

File: Main_filesize_vs_TurboJPE(...).png (21 KB, 900x556)

21 KB PNG

>>102105985

File: Bocchi the maid.jpg (375 KB, 1506x2048)

375 KB JPG

>>102105985
Bocchi the maid!

Also, why should I care about these over PNG?

>>102106620
>Also, why should I care about these over PNG?
are you serious?

>>102106620
Maid

>>102106620
>Also, why should I care about these over PNG?
Because AVIF gives you less than 1% of the original file size with almost no notable quality loss

File: Bocchi the maid2.jpg (397 KB, 1600x2338)

397 KB JPG

>>102106627
Yes. I enjoy using PNG via BufferedImage but could be persuaded to use a different format if there is a good reason to.

>>102106691
It is nice to see you Kirumi. How is translating things for /x/ going?

File: DecodeMPs_Chrome-110-Win_(...).png (30 KB, 900x556)

30 KB PNG

>>102105985

>>102106749
filesize.

File: 1724763541035.jpg (95 KB, 735x817)

95 KB JPG

>>102106749
Hi. Im making progress, its going well. What about you and the science foundation? Any news?

File: 1724763972638.jpg (224 KB, 1426x2048)

224 KB JPG

>>102105985
Nijika (my love) prefers JpegXL.

File: city.webm (555 KB, 400x400)

555 KB WEBM

time avifenc -q 80 city.png city.avif
 * Color total size: 2635896 bytes
 * Alpha total size: 0 bytes
Wrote AVIF: city.avif

real 0m1.581s
user 0m4.891s
sys 0m0.060s

time cjxl -e 5 -q 90 city.png city.jxl 
JPEG XL encoder v0.10.3 4a3b22d2 [AVX2,SSE4,SSE2]
Encoding [VarDCT, d1.000, effort: 5]
Compressed to 2557.4 kB (0.781 bpp).
4180 x 6264, 29.717 MP/s [29.72, 29.72], , 1 reps, 8 threads.

real 0m1.303s
user 0m4.325s
sys 0m0.153s

>>102106887
What am I supposed to see here?

>>102106949
That JewXL is slightly superior at visually lossless compression (terrible at everything else)

File: Cyan.gif (317 KB, 540x540)

317 KB GIF

>>102106748
If it is lossy then I don't want to use it. Lossy """"""""""compression"""""""""" is useless for my projects and I don't actually care about the image sizes of the files in my maidposting folder. If I can't guarantee that every RGBA value in every pixel will remain exactly as I have set it, then I have no use for the format and will not intentionally use it for any reason.

Are any of these offering file size decreases losslessly?

>>102106811
The hash counter is nearing completion and will be posted via KMCII into a thread later and then I will make another take a paper leave a paper thread with a differenct Deterrence Posture and see if janny respects my thread.

>>102106960
I don't see a difference in any of the 3 images. Like there are some pixels moving around but the quality of them is all the same.

>>102105985
jewxl bros/sisters, our reponse?

>>102107052
The difference is so small you have to expand/zoom in.
Basically JewXL keeps more grain present in the png whereas avif make it a flat surface. Some of the slabs/bricks around the windows also have slightly more defined lines on JXL.

>>102106887
JXLbros we are so fucking back.

>>102107139
If I have to zoom in to see a difference I don't care.

File: 1723012348821.webm (3.66 MB, 1920x1080)

3.66 MB WEBM

What the fuck does modular encoding in jpeg xl even do and why is it butchering outlines in my Any May Mays?

File: city.webm (495 KB, 400x400)

495 KB WEBM

>>102106960
>terrible at everything else
Here's the same test at medium quality.

time cjxl -d 3 -e 5 city.png city.jxl 
JPEG XL encoder v0.10.3 4a3b22d2 [AVX2,SSE4,SSE2]
Encoding [VarDCT, d3.000, effort: 5]
Compressed to 763.9 kB (0.233 bpp).
4180 x 6264,  41.272 MP/s [41.27, 41.27], , 1 reps, 8 threads.

real    0m1.050s
user    0m3.646s
sys    0m0.126s

time avifenc -q 60 city.png city.avif
Encoding with codec 'aom' speed [6], color quality [60 (Medium)], alpha quality [100 (Lossless)], tileRowsLog2 [0], tileColsLog2 [0], 8 worker thread(s), please wait...
Encoded successfully.
 * Color total size: 832434 bytes
 * Alpha total size: 0 bytes
Wrote AVIF: city.avif

real    0m1.274s
user    0m3.924s
sys    0m0.083s

>>102106960
Tel-AVIF shills truly know no limits to their attempts to deceive

>>102107424
>1 FPS
Stop doing this YOU DUMB FUCK.

ffmpeg -framerate 1/3 -i %%02d.png ^
-r 30 -c:v libvpx-vp9 -profile:v 1 -pix_fmt yuv444p ^
-color_range 1 -colorspace 1 -color_primaries 1 -color_trc 1 ^
-b:v 0 -crf 20 -quality best -speed 0 out.webm

>>102107435
Says the DSSIM cheaters butchering anime with pixilated outlines.

File: IMG_6723.jpg (555 KB, 1572x2004)

555 KB JPG

>>102105985
Bocchi would never use JXL.

File: chromium video player.png (195 KB, 394x396)

195 KB PNG

>>102107459
You can't be for real.

File: Bocchi the maid3.jpg (490 KB, 1032x1457)

490 KB JPG

>>102107488
Bocchi would never use lossy compression.

If I cut your feet off to make you shorter, I didn't compress you. I mutilated you.

File: 1712277396642050.jpg (3.87 MB, 4632x3088)

3.87 MB JPG

>>102107006
>Are any of these offering file size decreases losslessly?

cjxl --lossless_jpeg=1 ./maid.jpg ./maid.jxl
JPEG XL encoder v0.10.3 4a3b22d [AVX2,SSE2]
Encoding [JPEG, lossless transcode, effort: 7]
Compressed to 3082.3 kB including container

https://files.catbox.moe/eov29i.jxl

>>102107508
Have you even tried that? It still skips over the JXL frame at quarter speed.

>>102105985
Kinda crazy that AVIF looks almost lossless at just 0.5% of its original file size. It's almost like magic.

>>102107424
>speed 6
>no 10-bit color precision
>no yuv 444
>no cicp 1/13/1

I mean we all know the defaults are dogshit by now but modern CPUs can literally encode 10-bit 444 4K AVIF within seconds at speed 0 now. There's not much reason to use speed 6 anymore.

File: Bocchi the maid.png (1.4 MB, 1654x2339)

1.4 MB PNG

>>102107546
>3082.3 kB
What was the uncompressed filesize? What was the size of the input?

File: works.png (30 KB, 200x193)

30 KB PNG

>>102107549

>>102107459
>>102107549
What are you even talking about?
You're supposed to pause and seek between them.

>>102107570
Why are you so skilled at getting your jewish image codec to cheat in DSSIM but not at creating 30 FPS still webms?

>>102107570
>Daiz
>Try Not to be Insufferable Challenge: Impossible

>>102107563
>>no yuv 444

  Stream #0:0[0x1]: Video: av1 (libdav1d) (High) (av01 / 0x31307661), yuv444p(pc, smpte170m/bt709/iec61966-2-1), 4180x6264 [SAR 1:1 DAR 1045:1566], 1 fps, 1 tbr, 1 tbn (default)

>>102107580
I don't feel like it. If you're going to hack webms to use them for images at least display the frames for more than 0.00001 milliseconds you downie.

>>102107580
The container has built-in features to make viewing this exact type of content easier, and they are trivial to use.

>>102107582
>30 FPS still WebMs
Because that's fucking retarded. This is a (You) problem.

>>102107631
Your infamous stubbornness goes well beyond image format preferences, and bleeds into every facet of your personality. Interacting with you is torturous, and is likely much worse for those closest to you.

>>102107631
>"Just pause at the right millisecond goy!"

>>102107662
You're welcome.

>>102107669
PEBKAC

By the way, -crf 20? For a visual test?
You're retarded if you don't use lossless encoding.

>>102107718
CRF 20 video is used in production.

>>102107736
For video? These are still images where every single pixel matters.
Also, yuv444p is inherently lossy for JXL. Do you even understand the parameters you're using?

>>102107834
You don't need MUH LOSSLESS to see differences. You can still see them through CRF 20. Use CRF 10 if your autism is that bad but using CRF 20 with VP9 is like using quality 90 with JPG.

>>102107894
>Dude, just be sloppy who cares.
Yeah, no. Even a JPEG at 100% quality is insufficient due to irreversible color space conversions. 10-bit autism exists for a reason.

If you really must not use RGB for whatever reason, YCoCg-R is the way to go. I bet you've never even heard of that.

>>102107961
insufferable cunt

File: Screenshot_2024-08-27_16-43-00.png (40 KB, 552x467)

40 KB PNG

I hope this is the right thread to ask. Sometimes when I open an image in Gimp it asks me to convert the colors or some shit. What is that even supposed to mean and should I press keep or convert?

File: ebussy_laughing.webm (547 KB, 1024x1024)

547 KB WEBM

>>102108243
What's the usecase for a color profile?

>>102108268
>paid (((AOM))) promoter posting vile imagery
Not surprising.

>>102108067
I take that as a yes.

>>102108243
Probably yes. WebP has no business not being sRGB.

>>102108386
kek

>>102108514
So I always convert it no matter what? And it will always be lossless and not lose any color information?

File: dssims.png (61 KB, 1628x47)

61 KB PNG

>>102107424
Today I will remind you that you got BTFOd by webp
https://desuarchive.org/g/thread/102055186/#102073413

>>102105985
>>102106031
>>102106054
Smells like a synthetic, exaggerated comparison.
Are both sides taken from the same RAW footage? Compressed using the same values?

File: 1714659910873845.png (1.1 MB, 1920x1080)

1.1 MB PNG

>>102108921
>RAW anime screenshot
Literally what?
the PNG is LITERALLY linked in the OP. If you think you can do better, go ahead.
Protip: you can't. JXL is abysmal att high compression and line art, this well known and an expected result.

i'm gonna be hoenst as muhc as i love jxl avif absolutely demolishes it at low bitrate

>>102108959
No shit sherlock.
It's sad JonSneeder decides to be so dishonest shill rigging tests instead of simply conceding this.

>>102108959
We're supposed to expect the internet cares more about progressive loading than actually saving bandwidth. I would unironically not even open 80% of all 4chan images if they had a 720x720 AVIF thumbnail viewport. That alone would would probably reduce bandwidth usage by like 80% on average in this dumpster fire of a website.

>>102108985
cope and sneed

bocchi and nijika both have extremely hairy pussies

>>102109844
This is true and cannon btw

>>102108985
>instead of simply conceding this
That's rich.
I'm not the one constantly employing ad hoc rescues by dismissing everything unfavorable as rigged.
I can't concede on something I never disputed.

>>102109966
It's already an improvement that you stopped spamming these daiz schizo pastas in every thread

File: Tohru Question.png (1.28 MB, 1920x1080)

1.28 MB PNG

>>102107546
What size is it when converted to PNG?

>>102105985
Why wouldn't I just keep the original png? I don't give a shit about file size (I wouldn't convert actual 200MB photos to this, others are insignificant in size). I don't want to use lossy compression.

>>102110556
It's for sharing images on the web. Lossless is really le hecking cool and whatnot but websites have to pay bills and a lot of people around the world have data caps and thus are forced to survive on as little as 100 MB per day max so a 100KB AVIF vs a 1MB PNG is really appreciated by these people.

Also where can I find a phone with a 3.5 HDD slot?

>>102110699
It's more like 10kB vs 2MB

File: 0.2_BPP_comparison.png (2.04 MB, 3840x1080)

2.04 MB PNG

I think we're being brutally unfair to Jpeg XL only comparing it to AVIF at 0.1 BPP or lower so I've decided to crank it up all the way to 0.2 BPP even though this might be seen as a waste of bandwidth by most websites. "Give Jpeg XL a chance!" And whatnot. VMAF was used for this test because as it was mentioned in >>102102709 Jpeg XL has been caught cheating in SSIM/DSSIM. Modular was NOT used for the Jpeg XL of course.

92.90 VMAF speed 4 52.6KB AVIF
https://files.catbox.moe/41pww5.avif

91.47 VMAF effort 7 54.5KB JXL
https://files.catbox.moe/bdiq2a.jxl

I also used a non-anime image to be more fair to Jpeg XL. It honestly didn't do too bad but I would much rather have the AVIF with less visible artefacts present. VMAF seems to agree as well.

File: in.png (1.73 MB, 1920x1080)

1.73 MB PNG

Here is the original image used. On a related note using VMAF for images is going to be a whole new world of autism for us all but at least in my opinion scores of 95 or higher seem to be correlated with "visually lossless" depending on what display is used (ie tablet/TV/monitor). Anything between 90 and 95 appears "medium quality" and anything below 90 is just straight up dogshit quality.

>>102110699
I wouldn't know, I don't store stuff on phones.
If it's not meant for people storing photos then who cares? Why do I care what some corporation does to save money?

File: sunflower_vmaf.png (54 KB, 1920x1080)

54 KB PNG

>>102111117
>95 or higher seem to be correlated with "visually lossless"
Far from it. Even with video, a VMAF of 95 is easily achievable at Youtube-tier CRF values, and it shows. Picture related.

Video in question is from the Xiph samples.

>>102111517
VMAF was actually carefully mapped to human MOS scores which is why it's not easy to cheat in it with Jpeg XL's modular encoding. VMAF of 80 is already mapped to 4/5 MOS so going higher than 90 is already entering the "visually lossless" range for a lot of mobile devices and potentially even laptop computers. VMAF of 95+ is really only needed for people watching 480p videos on giant 30 inch 50 PPI CRT TVs from WW2.

That said these scores don't correlate with images because obviously this is a case of using a chainsaw to build furniture.

>>102105985
Does she know a lot about technolocchi?

File: 1695085760787928.png (1.6 MB, 1920x1080)

1.6 MB PNG

I use jxl for mpv screenshots

File: anime-yamanosusume.gif (106 KB, 220x220)

106 KB GIF

>>102111517
I mean THAT IS why AV1 is such a big fucking deal. It's achieving outrageously high VMAF scores above 4/5 MOS. The following bit-starved AV1 encode would EASILY get a "10/10 video!" YIFY score by like at least 80% of humans on this planet.

https://desuarchive.org/g/thread/101762605/

P5 CRF 50, 408 kbps
https://files.catbox.moe/2tyi6w.webm

File: sunflower_ssim.png (55 KB, 1920x1080)

55 KB PNG

>>102111618
>it's not easy to cheat in it
If your definition of cheating is the use of special encoding parameters, then it definitely is.
The AOM encoder can use a tuning mode which loads external libraries that optimize for VMAF at the expense of other metrics. I've tried it once, and it looked like oversharpened garbage in my humble opinion.
It also appears to be negatively correlated with SSIM as shown in this and the other graph. Go figure.

If you're curious to see the VMAF mode, I am willing to demonstrate it.

>>102111830
What ultimately matters is MOS doesn't dip below 4/5. OFC human perception can be trolled into thinking something is "high quality" when it really isn't.

It's like you said metric chasing will doom us all. If you can get people to go "10/10 video!" all the time with optimizations for good VMAF scores then that's good enough for the internet. Anything else is just us nerds looking at a screen 1 inch away trying to find unicorns nobody really cares about.

>>102111117
https://nitter.poast.org/jonsneyers/status/1573371624132419585#m

File: jonsneeders.png (104 KB, 300x300)

104 KB PNG

>>102112052
https://youtu.be/wA6sack7y3o

>>102105985
>left half avif
>right half jxl
>both sides are converted to ifg-bocchi.PNG

I think this is propaganda

>>102112052
>2022
Also not that shocking desu, regarding image quality VMAF scores under 90 don't make a lot of sense but at least in my own testing/opinion anything above 95 got well within the "visually lossless" range on my 1080p 16 inch ~140 PPI laptop anyway.

Would be nice if someone tested VMAF on a giant low PPI 4K TV.

>>102112140
Why is this jew so HIDEOUS? Can't he at least get rid of his black DNA afro and shave that satanic goat beard?

File: sneyers.jpg (1.25 MB, 4096x3492)

1.25 MB JPG

>>102112285
>jew
That's sugarcoating. Call him a B*lgian.

>>102112355
I mean that hair.


EWWWWWW.

File: jonsneyers.jpg (147 KB, 1920x1280)

147 KB JPG

>>102112367
It doesn't define him.

>>102112381
See that's much better. You can't immediately tell he has like 5% african DNA.