I finished the highres, high file size, high quality comparison between JPG, JXL, and AVIF. Image dimension is 5774 x 3929. PNG test file generated from the 130MB .tiff file found in:https://www.loc.gov/resource/highsm.54191/picrel is crop from the PNG test file.https://files.catbox.moe/61uh3m.pnghttps://files.catbox.moe/h3tv8n.jpghttps://files.catbox.moe/hn3z66.jxlhttps://files.catbox.moe/capka8.avif
Forgot: ALL IMAGES WERE ENCODED TO 2MB filesizepicrel is from JPG encode, colors look bad and artefacts are visible
picrel is from JXL encode, looks very identical to PNG source
picrel is from AVIF encode, looks very identical to PNG source as wellI don't get it, JXL doesn't appear to look better here at high quality. Is there something specific I'm supposed to be looking for?
meds
>>110034731Is the original file lossy or lossless? what kind of compression you're looking for? do you have a target file size?
>>110034709I appreciate the scientific method uploading pngs of compressed images, but literally what the fuck is the god damn point dude? look at the fucking source image, so much iso grain and smartphone fuzz, PNG is fine.
>>110034780The anons from another thread said that AVIF was not good for high quality/ high res image so I wanted to test this to see if it was true. It doesn't seem to be true from my subjective point of view so I'm wondering if there's some specific detail I should be looking for here.>>110034787Original is PNG generated from the .tiff linked in my OP, target file size is 2MB and desired and observed quality in both JXL and AVIF is high quality.
>>110034798That's not high res though. Try downloading something like NASAs bigass picture composites of galaxies and planets.
>>110034796Basically we know that AVIF is excellent at low res/file size/quality. What isn't understood is whether it is good at high res/file size/quality and that's what I'm testing here, my own subjective viewpoint is that JXL isn't better than AVIF at high res/file size/quality, they seem to be equally good. Could I be wrong though, is there some minor detail I'm not paying attention here?Here's a repost of the low res/file size/quality findings from the other thread:I tested the SFW Lenna image and chose 10KB file size as the control, the JPG version is obviously dogshit because it's an image codec from the 90s.https://en.wikipedia.org/wiki/Lenna#/media/File:Lenna_(test_image).png>>110034821You need lots of RAM to do that, I only have 16GB.
The 10KB JXL version looks much better than the JPG.https://files.catbox.moe/lcpvzh.jxl
but imho the 10KB AVIF version looks somewhat better than the JXL.https://files.catbox.moe/dwfh57.avif
>>110034834the detail youre missing is that all of this is entirely pointless
>>110034860You don't need to tell me that, pointlessness is the whole purpose of /g/.
>>110034834Download more ram bro
>>110034864Nah it has entertainment value
>>110034865I'm not stupid enough to buy a mac.
>>110034874Really?
>>110034709I don't see the difference. Try to find some objective tools.
>>110034731>>110034724>>110034717>>110034709if your source is a png, only lossless compression countsWe want to NOT LOSE QUALITY whatsoever.>but but it looks close enoughbut does it still look close enough after getting recompressed 3000 times in the same way?
>>110034709>high res>5774 x 3929Maybe in 1990.
>>110034709why is this even a discussion>jpg is old shit and shouldn't be used>avif is the web format, optimized for filesize>jxl is the everything format when it comes to raster imagesif you're a globohomo bigcorp whose pipeline optimizes images down to the last byte, you pick telavif. it's essentially a modern, less shitty webp, but it's designed and optimized with a heavy focus on filesize, especially at the extreme low end where visual degradation is acceptableif you're anybody else you pick jxl so you have an all-in-one formatsomething like your example where a target is a huge 2mb file is retarded, you won't be able to really tell them apart. avif pulls out if you were to target 400kb, jxl pulls ahead the closer to lossless you go.
>>110035007>teavifShalom!
>>110035007>jxl pulls ahead the closer to lossless you go.That's not what I'm personally seeing here though, I expected the JXL to look better, if that makes any sense.
>>110035001>if your source is a png, only lossless compression countsEven if it reduces filesize by like 90% and you have a hard time telling the source and lossy encode apart?
>>110034709If even one single pixel has a slightly different value, i consider your recompression faulty and broken.When i put a textfile into a zip file and extract it again, i expect the text to be exactly the same.I do not accept random typos as "close enough".Its a lossless PNG? Your compression has to be lossless as well.It is a lossy jpeg? Your compression has to be either lossless or a lossless transcode (only jxl can do that).
My attempt. I chose this image of Io.https://science.nasa.gov/photojournal/global-image-of-io-true-color/Original TIFF (17.281.563 bytes)PNG (OxiPNG + PNGGauntlet; 8.619.457 bytes)https://files.catbox.moe/oogens.pngJXL lossless (q100 and Effort 10; 4.909.885 bytes)https://files.catbox.moe/nd99jb.jxlJXL lossy (targeting 350MB; q71 and Effort 10; 346.775 bytes)https://files.catbox.moe/7efqpb.jxlAVIF lossy (targeting 350MB; q57 and Speed 0; 353.152 bytes)https://files.catbox.moe/4hwmkj.avif
>>110035043That doesn't even make any sense, comparing text to images is apple to orange comparison.
>>110035041>Even if it reduces filesize by like 90%Yes.If you advocate for recompression, you have to consider that this recompression may be done back and forward a thousand times.May i see your comparison of generational loss?If you didnt even consider this, then you are an idiot. Don't talk about subjects you do not understand.
>>110035001>but does it still look close enough after getting recompressed 3000 times in the same way?AVIF starts to oversaturate https://www.youtube.com/watch?v=w7UDJUCMTngIt's not too bad, but it's worse than JXL
>>110035052This would crash/freeze like 80% of computers. Images like these should remain PNG/JPG for now.
>>110035043 Jpegs at 100% quality are fuckhuge though, more than PNGs
what's wrong with png?
>>110035054Lets compare it to audio where everybody archives with FLACS?And when someone goes from mp3 to opus, he still gets the FLAC to create the opus from that one, rather than recompressing the mp3 to opus?How about binaries? Would you accept it if zip files flip random bits in the binary you compress?You advocate for a loss of quality.You are a subhuman.
>>110035059k
>>110035052Oh shi! The madman actually did it! I was joking around lol
>>110035067I encoded all these images on my old ass R7 1700x
>>110035063nice, so AVIF is not an option even when you are ok with losing quality.Then this subject is over, a conclusion reached.
>>110035063Also seems to blur faces https://www.youtube.com/watch?v=oYM9spW7VBQ
>>110035092I should have just taken screenshots so that people don't have to open youtube
>>110035080The resolution is smaller than the image OP picked. My choice was Io mostly because it's a very colorful moon
>>110035063
>>110035092Awesome!See how easy this was?No need to fuck around with made up metrics or idiotic "well, it looks identical".Just run it a thousand times and look at it.So AVIF and WEBP can get fucked.There is no argument against this. No away to jew around it.It's over.
>>110035110>>110035101tldr: jpeg = sovl, others = soulless
https://www.youtube.com/watch?v=8zdSKzLvsGA
>>110035073So everyone who listes/broadcasts FM radio should be sentences to death? Do you have any idea how deranged you sound right now? This reads like a truth social post by zognald desu.>>110035063>>110035092>>110035101>>110035110What a dumb pointless outdated "test" from 2020 that has nothing to do with my OP findings where I convert a PNG source to either a lossy JXL or lossy AVIF in just a single generation.
>>110035118>2020
>>110035101>>110035110>>110035118this is fascinating, but no one is going to re-encode the same image a thousand times
>>110035123>So everyone who listes/broadcasts FM radio should be sentences to death?yestotal boomer death
>>110035123do you record music from FM radio and store it as mp3 or opus?If you do that, then yes, what you do is retarded.
>>110035130>but no one is going to re-encode the same image a thousand timesclearly you haven't been around in the early webp era
>>110035127I propose you post up-to-date comparisons.
>>110035130>no one is going to re-encode the same image a thousand timesAre you sure about that?On the internet, where people save images that were recompressed on upload, then upload it again, where it may get recompressed again, and do this again and again?
Interesting. With text JXL causes color bleed while AVIF causes blurhttps://www.youtube.com/watch?v=mKMRzbuK9rA
>>110035130It's also from 2020 and we're all expected to assume like zognald supporters that nothing has changed since then.>>110035142If you retards intend to encourage more people to adopt JXL please don't use outdated shit, people will notice, and you will do more harm than good.
>>110035151>>110035123>but but its outdatedThen create a more recent test.You spend your whole life on shilling this garbage, but are incapable to test THE ONLY METRIC THAT MATTERS.
>>110035151unless you prove that current webp or avif improved their generational loss performance, i consider this discussion over.By objective visible and easily repruducible metrics, webp and avif are unsuitable as formats for sharing pictures on the internet.Its over.Done.
>>110035052forget lossy compression, jxl blows everything else on lossless
>>110035160More harm than good. You may think you are being slick and successfully astroturfing this way but people will notice and see that JXL supporters are very dishonest niggers.
>jxl: lossless support, lets you upcycle jpg with zero loss, can be used as payload in dng and whatnot, will still be supported by 2050>avif: slightly better at the ultra low end filesize, replaced in 5 years due to av2 being a thingwow I'm having to think about it very hard
>>110035176Do you disagree with the notion that preserving quality is the most important subject?
>>110035130>no one is going to re-encode the same image a thousand timesSocial media platforms willWhy do you think so many meme pics look so damn crusty?Last month I sent my mom a photo I took via Facebook Messenger (literal boomers pretty much only use Meta platforms), she then reposted it in our family groupThe image was re-encoded both times and it looked like dogshit after the second upload
SOOO, anyone want to go back on topic and point out what I should be looking for that points to JXL being better at high res/file size/quality image compression for AVIF for my OP F-35 lossy test JXL and AVIF images?Or do you also agree that JXL and AVIF are now tied in high res/file size/quality?
>>110035208It's so annoying that those cheapass telecom bitches refuse to build out a proper global internet infrastructure so filesizes barely matter. But nope still gotta listen to 128 encoded .ogg on soundcloud because 3rd worlders need their mumble rap too
>>110035215>point out what I should be looking forgenerational loss
>>110035222kek
>>110035215If you're a photographer you use raw anyway and that shit's huge
>>110035227>photographers use RAW anyway...which now may use jpeg-xl (lossless preferably), since it got adopted in the DNG standard
>>110035215Can you create a test where you compress the same pic over and over again?Since this is what happens to pictures shared on the internet, its important to me, when choosing a format for sharing on the internet.
>>110034709webp/jxl/umigalaxy schzo thread
>>110035219128 kbps opus is borderline transparent so i'm not really seeing the issue with this>muh golden ear muh 24bit/96kHz you're listening to music on a $200 chinkphone or a used stinkpad, calm down lil bro
>>110035241OH THANK GOD, I thought you were going to accuse me of being "daiz".
>>110035241It's just Daiz. He's multitasking sperging about Malfoid and JXL.
>>110035230You have no idea how stubburn artfags are at adopting new technologies are. If their outdated software doesn't support it, they won't use it.
>>110035219
>>110035242The issue is it's auto encoded yo shit completely mangles the sound. Although lately it's been getting better. Mostly because competent people that make sound work around the limitations.
>>110034709I wanted to try if an AI could figure out a more efficient image predictor algorithm to compress the image even further https://www.januschka.com/jxl-art/Sadly there's a file size limit.
>>110035242nta, but take a FLAC and convert it to opus yourself and compare its quality to whatever spotify or soundcloud or any streaming platform for normies are serving.The difference is crazy.>chink phonea 50 bucks in-ear headset on any random cheapo built-in DAC is enough to clearly notice the difference, unless you are deaf.Normie platforms do not deliver what they claim to deliver. Spotify claims to do opus, but what most people get served by it is ogg vorbis or acc.This jew-behaviour may have caused the other anon to believe that opus 128k would be bad.
>>110035252All Apple shit supports it, and artfags love to be special with their Apple shit.
>>110035303Yeah I bet they're gonna run to the store and buy the latest 5k Canon camera house just for a file format with no real benefits. Just to please some anal retentive autist on 4chan.
>>110035325>but someone may have old hardwareAnd? Are you in a hurry?Most new art isn't in HDR yet either. But it will be in the future.
>>110035325>with no real benefitssorry, i just skipped over this.. turns out that you are just dumbHDR is very valuable to artfags, because they love their amazing-color-explosion displays.What choices do you have for lossless HDR?Huh?What format would that possibly be?Now imagine that this is the same format as its in the DNG raw already.Artfags are eating good here. They get out of the hellhole that is proprietary RAW garbage (at least partially) and can get rid of that disgusting HEIF, which they never liked in the first place.
Does libjxl still use a fuckton of RAM when encoding high res images? That made me drop JXL a year ago but might give it another try if that's been fixed.
>uh jpg is bad because you can't recompress it a million timesJust save images instead of sceenshotting them. Is that really so hard?
>>110035449If nobody ever recompresses anything, than this thread is pointless.
I recompress all my frogs to lossless so the bits don't degrade over time
>>110035479this thread is pointless and OP is an LLM bot
>>110035479Yeah and maybe we should just compress "then" and "than" into the same word to save you the trouble of trying to figure out how to tell the difference between them.
>tell ai sloppa to not do $thing unless explicitly asked>it does $thing on its own
