What does /g/ think about AVIF?

It's great for mpv screenshots. picrel is from 853x480 video scaled to 4K, the source AVIF was only 149 KiB. Looks indistinguishable from playback.

it sucks

>>105559841
>t. daiz

>>105559133 (OP)
it's great, just slow to encode.

>>105560050
-s 6 is really fast and still efficient, usually takes less than a second.

Low blocky quality. Also you made this same thread earlier.

>>105559133 (OP)
jxl is obviously better

>>105559133 (OP)
for lossless: jxl > png > avif.
for lossy: jxl > jpg > avif.
why? colour accuracy. avif sucks so much at that.
i don't even consider heic and webp, the former has licence issues, the latter has quality issues.
2kliksphilip has a video about this: https://www.youtube.com/watch?v=SzsM4HMKmEI

>it's another episode of Daiz and Jay manipulating facts
grim

>>105559133 (OP)
I think it's dogshit at lossless encoding (barely any better than png and absolutely terrible compared to webp), and not good enough at lossy to justify using over mozjpeg

>>105560917
>colour accuracy
>avif sucks so much at that.
>https://www.youtube.com/watch?v=SzsM4HMKmEI

>sample size: 1

I can also pull up one image where it fares better than JXL in that scenario. Just don't compress it that hard if you care how it looks under zoom.

Avif won.

It's better at being WebP than WebP.

>>105560917
Similar shit quality settings as the fagtuber, totally different outcome in terms of color. JXL's still fucking awesome and there's no need to be a disingenuous nigger.
>for lossy: jxl > jpg > avif.
lol

>>105560435
I've replaced jpg user uploads for avif and the processing queue would sometimes take ~20s with batches of 10 images.

>>105561771
trvthnvke

>>105559841
>>105560630
>>105560917

Here's an actual relevant HIGH QUALITY comparison with ssimulara2 scores/settings/etc I saved from a thread a while back. Not anime but not a real life photograph, making it a decent middle-ground test for image codecs in general. It would appear at least from my limited research that AVIF used to be dogshit and JXL was somewhat better in terms of general compression efficiency taking both photographic and non-photographic content into account. However something change in the past few years where AVIF has achieved parity with JXL on average.

I have to stress that ON AVERAGE there doesn't seem to be a "winner" in terms of compression efficiency, using JXL or AVIF will probably get you good results. With that said AVIF seems to performs better with anime content and JXL performs better with photographic content. Anyone who tells you otherwise is fucking trolling. Where JXL is arguably better is that it launched with progressive decoding day #1 whereas AVIF only recently got that so if slow connections are a concern then JXL has an advantage there unless you explicitly go out of your way to make your AVIF images progressive. Which to be fair at file sizes of just a few hundred KBs doesn't really seem necessary.

>>105562124
5MP pics take half a second >>105562076 I don't know what to tell you

>>105562076
Whoops, sorry I overlooked your image anon. Very bizarre photographic result I can't explain, is it because you used 12-bit depth? Also what the fuck is the -a tune=iq option? I've never seen that before.

>>105562289
does avif still cap at 2160p?

>>105562401
That's a tricky question to answer and the best I can find online is "+1
AVIF, or AV1 Image File Format, supports a maximum resolution of 65,536 x 65,536 pixels. While this theoretically allows for incredibly high resolutions, the Wikipedia specifies that AVIF Baseline profile, designed for broad compatibility, limits individual coded image items to 8K resolution (8192 x 4352 pixels) to ensure wider decoder support. Larger images can be created using grid derivation, but this might introduce artifacts at tile boundaries."

So it appears that the MINIMUM "limit" is 8K x 4K in resolution but AVIF can hack together larger images together by making maps of images in a single AVIF file.

>>105562401
Also it's not clear how bad those "artifacts at tile boundaries" might be. With that said this is a limitation present in the BASELINE profile meant to ensure maximum compatibility so it does seem like you could just make a 65k x 65k resolution AVIF without using grid derivation but compatibility would obviously be a concern.

>>105562436
>>105562467
ty

>>105561403
/thread

dead on arrival pointless format

>>105562396
>is it because you used 12-bit depth
JXL has even higher bit depth (16), AVIF is at a disadvantage if anything.
>what the fuck is the -a tune=iq
SSIMULACRA2 tune

>>105562289
>With that said AVIF seems to performs better with anime content and JXL performs better with photographic content.
Pretty much, yeah. AVIF will essentially roundhouse kick JXL into a volcano at the speed of sound compressing anime. Like it's not even a competition.

>>105562490
I'm now going to bump this thread until someone takes my keyboard from my cold dead hands. Not because I disagree with you but because you threaded your own post and you must suffer for that gravely.

>>105562491
I can't find any information on human vision exceeding BT.2020 color primaries. I'm not sure what benefit there even is in storing 666 trillion trillion trillion colors. Like would that help better detect tumors on an MRI machine or something?

>>105562618
>AVIF will essentially roundhouse kick JXL into a volcano at the speed of sound compressing anime
I wouldn't go that far, most of the time AVIF pulls ahead it's not by much.

>>105562751
When you use a quality parameter of 50 and a speed preset of 0 with AVIF, JXL will AT THE ABSOLUTE BARE MINIMUM require twice the file size to match the quality with clean anime screenshots. That said this is the low quality side of things and as you up the quality parameter for AVIF, the humiliation for JXL is reduced. It's just something about AVIF and large areas of uniform color that turns it into a demon of an image format.

That's why I made a thread yesterday where if AVIF were allowed on 4chan I'd go out of my way to make threads filled with dozens of anime screenshots encoded with speed preset 0 AVIF and since quality 50 is more of a meme thing I'd bump that up to 69 as well. Like I mean this IS a website centered around Anime and whatnot so it seems kind of absurd that AVIF isn't supported yet.

>>105562268
sorry for preferring using an image codec for images instead of a video codec

>>105562618
>>105562899
>speed 0
It is excruciatingly slow.
>q 50
If I ever see such shit posted here I'm shift+clicking it on sight.

>>105563109
Speed 0 is fine for the 480p/720p anime screenshots I'm planning on taking. 10 years from now we'll probably be laughing how it only takes 0.1 second to encode at speed 0 on a computer from the future with an i9-1488 or whatever.

You know what I always find hilarious? Retards like @105563080 will always show up vomitting this same recycled dumb shit while at the same time not realizing that JPG by definition is also a "video codec". The doublethink in threads like these is astonishing.

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

>>105561600
ok goolag.
sample 2: https://www.youtube.com/watch?v=w7UDJUCMTng
sample 3: https://www.youtube.com/watch?v=mKMRzbuK9rA
sample 4: https://www.youtube.com/watch?v=S2SX1_xk2Fg
sample 5: https://www.youtube.com/watch?v=8zdSKzLvsGA
sample 6: https://www.youtube.com/watch?v=oYM9spW7VBQ

>>105562268
>avif is good because muh 666 gorillion dollar company said so

>>105563080
Videos are just images played in succession

>>105564093
That's very antisemitic of you anon, don't forget 6 million died for JPG!

>>105564066
You're obviously here to troll and therefore won't even acknowledge all the proof posted in this thread. What happened to this pointless dumb test in the webm format? You only spammed it like 6 million times so how hard can it be to recover it from the archives?

>>105564066
>misses the point completely

>>105564231
Anon if you're not worshipping jpeg xl as your personal god then you're literally Hitler. Think of the 11 million who perished! If you don't stop this immediately I'll be forced to shut down your bank account, make you homeless, and cancel you on social media!

>>105564229
>>105564231
>>105564268
I'm not daiz.

>>105564499
OK Daiz

Anyway I truly believe that the biggest obstacle to getting widespread adoption of AVIF is encoding speed and more importantly quality. Plebs like you and me can dick around with slow speed 0 AVIF CPU encodes and because we only have a couple hundred images on average and most of us aren't kikes who will have recurring nightmares from blowing $1 (if even that?) on electricity then none of this really matters to us.

However most websites ARE kikes and this is where things get complicated. They COULD just say "fuck it" and use speed 13 but the quality will be so horrendous that on average it'll only save like 10% compared to JPG lol. This is where hardware acceleration will be critical and it already exists but it costs $10,000 (ie picrel) and the KIKE websites obviously aren't too happy about this either.

Lastly it's fair to admit that hardware acceleration isn't a magic bullet. The first iterations MAYBE especially FPGAs (like picrel) could be achieving parity with speed 8 and this is being like really optimistic mind you. Depending on the quality level selected this could be reducing file size compared to JPG by like 40% (on average) instead of the usual 60% (again on average) speed 0 is capable of. This would still be a huge net benefit to the internet in general

For anyone curious, this is the $10,000 FPGA required to encode about 1488 4K resolution AVIF images per minute. If any richfag autist has that kind of money burning a hole in this pocket consider buying this thing and showing us if it does in fact achieve speed 8 quality compression.

>>105563452
That's as absurd as going to Japan and complaining about seeing Nazi symbols on their Buddhist templ-
Oh, that happened.

>>105564727
>using an amd gpu for hardware accelerated encoding
oh so that's why they don't declare what the efficiency is compared to a software solution... it's gonna be worse than jpeg lmao
>If any richfag autist has that kind of money burning a hole in this pocket consider buying this thing and showing us if it does in fact achieve speed 8 quality compression.
I could encode an avif keyframe with the av1 encoder from my rx 7600 and it would be the same thing probably, but no point in even trying it out, we all know it's going to be laughably bad considering that it barely performs on par with x264 slow

>>105564778
A "video codec" parroted by the village idiots of /g/ just have 1 simple requirement: YCbCr

Webp and AVIF both satisfy this requirement and they are therefore "video codecs" but GOLLY /G/ what about... JPG???

>"JPEG supports various color profiles, allowing users to specify how colors are represented in an image. The most common color space used in JPEG is YCbCr, which separates brightness and color information. When saving or exporting a JPEG, you may have the option to embed a specific color profile. This can be essential when maintaining color accuracy is crucial, especially in professional photography or graphic design where adherence to specific color standards is important."

https://www.lenovo.com/us/en/glossary/jpeg/

>>105564931
Incorrect, that $10,000 U250 is an FPGA. It is NOT an ASIC (ie the one in your rx 7600). Why do you think it costs $10,000?

Basically there are 2 popular flavors of "hardware acceleration". The quick and dirty is and ASIC while the FPGA is somewhat slower and okay in quality. An FPGA will never achieve the same quality as slow CPU encode but it will RAPE an ASIC in quality.

>>105564931
>I could encode an avif keyframe with the av1 encoder from my rx 7600 and it would be the same thing probably
So I went ahead and got this lossless webp (https://wallpapercave.com/wp/wp13865998.png) converted to this avif through amf_av1 (https://files.catbox.moe/uiaik0.avif) and this equivalent size jpeg through mozjpg (https://files.catbox.moe/uiaik0.avif)
the reason the result is 1920x1082 instead of 1920x1080 is that amd is so bad at making hardware encoders that this is literally an hardware defect that cannot be fixed in software, it just cannot output certain resolutions without adding black padding
overall, yeah it's shit the colors are all shifted and it doesn't really look any better than the jpeg
>>105565017
>Incorrect, that $10,000 U250 is an FPGA. It is NOT an ASIC (ie the one in your rx 7600). Why do you think it costs $10,000?
why do you think amd is capable of making an fpga that rivals hardware encoding for avif files when they historically suck so much balls at making hardware encoders? all these xilinx cards have going for it is the speed, that's why it costs 10,000$, the problem is that they NEVER ever talk about the encoding efficiency because then you'd realize it's completely pointless
like when they made that av1 encoder card 3/4 years ago that they said could encode like 100 simultaneous av1 high res high frame rate streams or something like that? yeah I wonder why no major corporation has deployed those cards, and I wonder why efficiency compared to existing encoders was never ever mentioned but they only ever talked about speed, weird.
my point remains: if avif is barely more efficient than jpeg when encoding through software, how the fuck can you expect it to make sense through an fpga?

>>105565096
>and this equivalent size jpeg through mozjpg (https://files.catbox.moe/uiaik0.avif)
*https://files.catbox.moe/snoa37.jpg

>>105565096
>if avif is barely more efficient than jpeg when encoding through software, how the fuck can you expect it to make sense through an fpga?
Prove it. Convert this 4K image to a 230KB JPG. Remember it must achieve a SSIMULACRA2 score of 80 as well. AVIF had no problem doing this as shown in >>105562289

https://files.catbox.moe/1y2x42.png

Also go watch a video on youtube between the difference between an FPGA and an ASIC. That might better help you understand the difference.

>>105565017
>YCbCr
So what are HDR videos encoded in ICtCp?

>>105565215
what kind of dogshit jpeg encoder from 1985 are you using

>>105565415
You can't worm your way out of this one. Either we drop this looney YUV requirement for an image codec to be considered a "video codec" or we admit that JPG is a disgusting Untermensch video codec.

>>105565477
????

>>105565521
>right click
>save
it's correct, his one is actually only 50 kb (probably because cloudflare transcoded it losslessly through mozjpg, further proof that he used some extremely poor jpeg encoder).
this is nothing new, these avif-tards do it all the time
they compare aom-av1 at the slowest preset possible which takes 5 weeks to encode to the fastest shittiest jpeg encoder they could find to make jpeg look bad when in reality it really is barely any worse than avif despite being 40 years older and way less complex/compatible, it's absurd.
just like that frahunofer graph that keeps being posted which claims hevc is more efficient than av1, it's all nonsense.

>>105565505
Who says that was a requirement? The distinction comes from taking a holistic approach. The chronology and codec functionality should be taken into account. In the case of AVIF, it was derived from the AV1 spec, which in turn leverages interframe compression. Ergo, it's categorically a video codec. Motion JPEG does not meet these criteria.

>>105565477
>>105565576
Nice try kike but the black sun let me see through your jewry.

>pix_fmt=yuvj420p

>>105565611
cope

>>105565636
>pix_fmt=yuvj420p
...so? does it achieve that simmulcra score or not? pixel formats don't fucking matter but you are all autistic about it for god knows what reason
>see? avif performs better than jpeg at encoding this extremely niche pixel format nobody has ever used or needed!
kys
you must be one of those retards that won't use opus for audio "because it resamples 44.1 khz to 48 khz and that's le bad!!!"
wake the fuck up, we are talking about LOSSY encoding here, if the colors get shifted by an amount impossible to detect by human perception, why the fuck does that matter?
>jewry
as if avif wasn't made by jews to save bandwidth by further reducing image quality kekw

>>105565636
What a compelling display of logos. I truly have been seduced by the might of such masterfully crafted rhetoric.

>>105565692
I'm not going to read all that but the whole point of shilling AVIF is that it's no longer limited to 420 blaze it.

The AVIF in this comparison is 444 so NO, you already failed to achieve SSIMULACRA2 of 80 hy default since thought you could coin clip us with 420.

>>105565748
>I'm not going to read all that
so you can spend HOURS daily making all of this nonsense comparisons and dozens of replies on these pointless threads, but reading like 4 lines of text is too much
sure, just admit that you have no way of counterarguing that faggot

>>105565803
Did you expect anything else from someone posting Nazi imagery?

>>105559133 (OP)
What does Hiro think?

>>105565803
Just post the 444 version already you coin clipper. Stop being a coward and get in the showers already.

>>105565835
>hiro
>think
That in itself is wishful thinking. We should be due for another hack in like a month at most.

>>105565864
kek

>>105559133 (OP)
Image formats that are just a video frame in a new container are inherently shit

>>105565692
Most JPGs on the internet use full chroma resolution you dumb fuck.

>>105565874
Like JPG? see >>105565017

>>105565850
>Hiel Hitler
You had one job

>>105566014
Obviously this wasn't made by an aryan. Still a useful idiot is a useful idiot.

>>105565876
>4k resolution 250 kb jpegs on the internet use full chrome resolution
you must be using a different internet than me then

>>105566060
It is a known fact that aryans are genetically literate

AVIF is a nazi image format apparently.