I haven't found a good encoder, I used all the ones with ffmpeg, such as libsvtav1 and librav1e, but the results are not as good. I screenshot the frame from the original and the transcoded file, then compare that with h265 and the av1 version has noticeable artifacts, like an object will move across the sky, and it will leave a few pixels behind sometimes. I just assume that the problem is with the encoder, and if I can get a hold of a GPU that does encoding, then I can finally start using it. Until then, I'm using vp9 for the short clips I publish, since that works on Discord and Telegram. Now if only Mastodon would support literally anything except h264/aac then I wouldn't have to add a h264 version to the pipeline for everything.
this post was submitted on 18 Jul 2023
34 points (100.0% liked)
Furry Technologists
1315 readers
3 users here now
Science, Technology, and pawbs
founded 2 years ago
MODERATORS
Hardware encoding is worse quality than software encoding iirc (per filesize). libaom should be the best quality codec but it encodes the slowest I believe? Idk the last time I checked on AV1 was like a couple years ago and the encoders are always in flux. You can use vmaf for a deterministic scoring system for quality if you don't want to pixel hunt.
What settings have you been using? I haven't noticed any issues so long as I'm not concerned with real-time encoding. And yeah, GPU encoding is generally worse than software, it's just usually way faster.
EDIT: for reference I've been using speed 6 and an RF between 40 and 30, and even in fast-paced scenes like ones in "Puss in Boots: The Last Wish" I can't notice anything super off. With my real-time recordings best I can do is speed 7 and 6000kbps (maybe higher bitrate), which isn't quite enough for the 1080p60 fast, colorful gameplay of Splatoon 3 - but even then, I'd need a decently higher bitrate with either x264 or x265, especially GPU encoded.
IDK his exact settings, but a fur I follow on Mastodon has been able to get 4K Blu-ray rips encoded to AV1 down to a handful of gigabytes, and he reports no noticeable quality problems at speed 6 RF 30.
The settings depends on what I'm doing. If I'm trying to squeeze a lot of video to a small space, I'll crank up crf. For good quality I was using crf 40 or so. If something is already encoded, simply transcoding will be lossy and be blurrier if you side-by-side them, so for archiving I decided to just always keep the source, even if it's raw DVD or other wasteful codec. But for making clips and things out of it, I'm trying to use the best thing, that doesn't waste a lot of space, and is playable by most people. It was awhile ago that I compared encoders, but I settled on using libsvtav1. librav1e seemed to have even less options, and it didn't seem better, but maybe there is a way to tune it better. What killed it for me was when I had an artifact in an AV1 video that I noticed, and when I encoded the same video with h265 and with same bitrate, it did not have that problem. Eventually one of my subscribers complained they couldn't view my h265 videos on their phone, so I switched to vp9 for more compatibility, and without having to stoop back to the least common denominator codec that is h264. So now even if I figure out the best encoder library and settings, I can't use it right now because of compatibility, I'm not just making videos for me to look at. I'll probably try pushing for AV1 again soon just to see, because it's inevitable that it will be supported everywhere. I'm sad to hear people saying GPU encoding can't be better, I hoped there would be an option to make it take longer but do good job at really great quality per byte. I still dont have a GPU that can do AV1 encoding, and I can't find any articles that show comparisons on the quality of the output, it's always just about how fast it encodes.
For hardware encoding, it will depend on your hardware encoder. Even for h264 etc, the hardware encoders have always been a step down from software encoding. This is a quick chart of Intel's AV1 encoder in comparison to other common software and hardware encoders. It's from this video which iirc was very informative on this topic for the Intel GPUs specifically.
I see they put out at least a couple more videos on AV1 in general here and here if you're interested, but I haven't watched them.
Thank you, that was very informative!
okay... so I'll try and get the usual complaints out of the way by saying I'm aware i'm likely talking about a >6 month processing time here, and i'm not saving any money. This is mostly for my amusement.
I'm looking to see if I can find a good one-size-fits-all conservative AV1 encoder+settings for my content, something that won't cause issues on content with fast motion that I can try to apply to everything in my library over time. I plan to use CPU encoding, and I plan to deploy this as a kubernetes service, with pods on most of my hardware.
basically I was curious what your best starting point would be for encoder and settings, since I'm a novice in video. (only played with handbrake so far). This is really just an excuse to play with kubernetes.
so long as my PLEX server can decode AV1 with quicksync i'm not too worried about hardware encoding. It can take months to encode and that wouldn't really bother me that much. I have an i5-13600k in my NAS with the iGPU shared to my plex container, so that SHOULD suffice.
PS: I promise I won't run any of this pod on the server handling pawb.social ~<3
I, too, have only been using Handbrake because idk what else to use, so I've been limited to SVT-AV1 for encoding. I'd need to watch through stuff at length, but at speed 6 and quality 40 I personally don't notice anything super off in my 1080p Blu-ray rip of The Last Wish (and got it down to an incredible 1.5GB). I definitely didn't catch anything at speed 6 and quality 30 - it won't be as small, and it'll still take a while, but at that I personally saw like. no problems whatsoever.
Might need to watch out for any film grain though, idk how much that mucks with the process but there are settings to denoise and reapply on playback. I haven't experimented with them too much because a) handbrake, and b) was trying to find good settings on a movie without grain to start.
I mostly saw problems when I used AOM-AV1 for encoding, at speed 7 and 6000kbps. That was purely because real-time encoding was why I needed that and it just wasn't quite perfect for 1080p60 Splatoon clips, mostly also as a test for "if Twitch turned AV1 on tomorrow what could I get at their current limits." SVT instantly got overloaded for real time at any real-time speed for some reason.
From what I saw on a chart from NVIDIA's website only the 4XXX RTX series even has hardware support for AV1 encoding, since I happened to be looking into the exact same thing last week lol. It does look amazing and should save so much space and bandwidth in the future!
Yeah, hardware encoding is very new, but you can get an Intel Arc 700-series card and get it via QuickSync for pretty cheap if you have a spare x16 slot and really, really need it right now. Only those, Nvidia's 4000 series, and AMD's 7000 series support encode. 3000 and 6000 series for each company support hardware decode, though.
Just need to get phones on the hardware decode train! Qualcomm and Apple, step up your game!
view more: next ›