What went right?
another episode of "AVIF Evangelists vs AVIF Detractors"
I don't really care about file sizes in the year 2024Which one will let me sear my eyeballs with blinding HDR?
Yay!! the hourly Avif shill thread!
>>101543869This. Wait for MUH hardware acceleration you cocksucking shills. The rest of us are sticking with JPG/PNG until they pry it from our cold dead hands.
Why does AVIF darken the image when using 10-bit?>>101516006As long as this is true, it's completely unacceptable for general use no matter how high the quality is.
>>101544104I have no clue what I'm supposed to be looking for here. Those images look pretty similar to me.
>>101544193Here is the 10-bit AVIF decoded with avifdec -d 8 just to keep the format consistent. It still looks darker to me.
>>101544104>>101544403I don't really see it. If I zoom in 400% I can see that lossy compression was used but it's not super obvious like it is with jpeg xl.I'm not sure what you were expecting though, lossy image formats end up discarding information 90% of people don't really care about to save space.If you truly wanted absolutely nothing less than 100% image quality then I guess you would use lossless compresion of jpeg xl? I don't really mind losing a small amount of quality if it saves buttfuck tons of space though IMHO.
Animations, AV1 video screencaps, and thumbnails. That's about it, however.
>>101545049see pic above
>>101545028If you can't see it, you can still measure it. GIMP's histogram shows a difference in median brightness, and the 8-bit AVIF scores a better PSNR despite being smaller. This is much worse than something you'd have to zoom into.It's not like the image is hypercompressed either. 85 is pretty high quality.
>>101545137Hmm I have a 10-bit LED gaming panel. That might have something to do with it. I get less color banding with 10-bit AVIF than I do with 8-bit AVIF so I'll probably still avoid using 8-bit.Maybe dithering 10-bit content to display on 8-bit panels will never be a perfect process.
Testingavifenc -d 10 -s 0 -q 93 --cicp 1/13/1,avifdec -d 8
avifenc -d 10 -s 0 -q 93 --cicp 1/13/1,avifdec -d 8
>>101545306cjxl -d 1 -e 9,djxl
cjxl -d 1 -e 9,djxl
>>101545317>>101545306Original from prior thread.
>>101545317file size?
>>101545359JXL: 363 KiBAVIF: 372 KiBAs shown earlier, there's some weird thing with the color going on in the AVIF. Banding is a lot more prominent.
>>101545306is that skyrim?
>>101545399even better
>>101545317I still see blocky artefacts on this one.
>>101545391>>101545317>>101545306Contrast turned up by "100" in GIMP.AVIF on the left and JXL on the right.
>>101545306>>101545391Q93 is a pretty ultra overkill setting for AVIF images. The 1-100 scale is actually mapped to 0-63 AV1 quant so "quality 75" would actually be quant 16 which few people outside of production would be interested in especially since diminishing returns begin at quant 20 which is like quality 70?Default is quality 50 or quant 32 but this only pertains to cellphone displays since such high PPI hides a lot of artefacting.
>>101545626Video quality demands are a lot more forgiving, so it's no surprise that the CRF targets are in the 20's.The avifenc quality scale instead tries to match JPEG, where 90 is a common target, even being the default in GIMP.
>>101545756Source? All I see are references to the quality setting being mapped to the quantisizer in AVIF so anything above quality 70 would only be useful for production not for user image distribution and this includes low PPI displays like desktop monitors.https://github.com/ImageMagick/ImageMagick/discussions/5371
One thing that seems very consistent with JXL VS AVIF is how they each choose to degrade the image.AVIF prioritizes pixel-perfect details at the cost of gradient integrity. If I lean back on my chair, JXL looks better. If I move my face all the way to the screen, AVIF looks better.
>>101546007It's the file size that really matters in the end and hands down AVIF absolutely fucking dominates below 100 KB for 4K images. There's 0 doubt about it. I'm not sure how many websites need this level of bandwidth saving though.4k AVIF under 100KBhttps://files.catbox.moe/ycknht.avifOriginal 4K PNG image: ~8 MBhttps://highdefdiscnews.com/wp-content/uploads/2020/05/the_invisible_man_2020_4k_37.pngAVIF CRF 40 (LQ): ~13 KBhttps://files.catbox.moe/f7epbe.avifAVIF CRF 30 (MQ): ~24 KBhttps://files.catbox.moe/l9i077.avifAVIF CRF 20 (HQ): ~42 KBhttps://files.catbox.moe/tdv8ee.avif
>>101545899Never mind, I am probably mixing it up with the JXL quality parameter, which states the following:>-q QUALITY, --quality=QUALITY>Quality setting, higher value = higher quality. This is internally mapped to --distance.>100 = mathematically lossless. 90 = visually lossless.>Quality values roughly match libjpeg quality.
Can't post it here, doesn't matter.webp/avif/jpxl/jp2k/ang/mng/vp9/h264/h265/h266/av1don't care, all useless since they can't be posted.jpeg/png/gif/vp8 4eva.
So.. What format is better for large scale bandwith usage? JxL?
>>101546146We got VP9 after that got enough hardware acceleration support. Probably same thing will happen with AVIF since the first 4K image posted in >>101546113 will donkey kick like 50% of old android phones in the balls.
>>101546169Depends on the level of bandwidth savings. Need 4K images under 100KB? I don't know what other image format besides AVIF is capable of such a feat.Maybe AVIF2 but that's like a decade away at least.
Chatgpt gave me this, hope it's correct: quant = 63 - (.63 x quality)Quality 50 = 63 - 31.5 = quant ~32Quality 60 = 63 - 37.8 = quant ~25Quality 70 = 63 - 44.1 = quant ~19Quality 80 = 63 - 50.4 = quant ~13
>>101546113Your "HQ" AVIF is SSIMULACRA2 score 82, Butteraugli 3-norm score 2.3. That's medium quality at best and will have visible degradation. Okay for some quick preview, but less than desirable if you actually want a high-quality image.
>>101546695Has anyone tried encoding AVIF with butteraugli tuning? It did wonders to my AV1 video.
