I must admit that I always liked the idea of free open formats, and AV1 was the only one that I took in consideration for my projects - avoiding Daala, VP8/9 etc. mostly because of the small base of potential users.
And put AV1 apart, too, because it was slooow to encode, difficult to decode, quality was not that great in comparison to h265 at same bitrate, and still small potential user base.
But lately I started to think back to it, made my homeworks, and found some interesting facts:
- encoding now is faster than h265, with better quality at same bitrate
- at same quality, AV1 file is at least 10/20% smaller than h265
- it could be easily decoded via software by average computers
- it could be hardware encoded/decoded by some GPU of last generation
- it could be hardware decoded by few cheap TV BOX/media players
- Netflix switched to AV1 recently
I've made few tests lately, and found that an UHD scope source could be encoded at around 60mbps preserving the grain perfectly!
Now, I'm curious if someone here is an AV1 expert, to help to write "perfect settings" for it, as documentation is scarce and there is no official physical media AFAIK that use it and that could be used as example.
EDIT
Made several encoding using only svt-av1 (Netflix and Intel contributed to its development); to have a common ground between encoding, I set the bitrate to a fair 50mbps - the source is from a 77mbps AVC
[email protected] version of the DCP trailer that may be up to 250mbps. Bitrate chosen because it's the minimum found on UltraHD Blu-ray releases, even if it could be lowered to 35mbps to get the same quality.
At first, based on my previous experience with AVC/HEVC, I've chosen an average speed of 4 (from 1 the slower to 8 the fastest) and dual pass; it took about 147m to encode (so 63x real time); result is very close to the source:
https://screenshotcomparison.com/comparison/19862
Then I've discovered that speed 8 strangely has more quality than speed 4 (
graph source):
so the second encode was made at speed 8, always 2 pass; this time encoding time was only 47m (20x real time) and quality the same; so, no reason to use slower speed (but according to the graph, speed 7 should retain a bit more quality with a small speed time tradeoff; must try it in the future).
Then, searching for the best settings, I've found
a post on reddit about grain synthesis: let give it a try, then!
Again same encoding settings of the last one, plus grain 15; more or less the same encoding time, but with a nice grain that covers those ugly encoding artifacts found on the source (that is, actually, a mere AVC encoding of the original that is, again, an encoded version); so, at the end, a third generation version that looks like better than the second one IMHO!
Still... let's try the same, but this time only one pass; encoding time reduced to only 27m (11.5x real time) using one thread and 16m38s (7x real time) using 16 threads! Quite a long encoding time, but still "only" 14 hours to encode a 2 hours long movie, instead of five using speed4 2pass, and also better I must admit! And differences between 1pass and 2pass were negligible, and only possible to be seen when zoomed in at 800%, so I dare everyone in the world to see the difference while playing!
https://screenshotcomparison.com/comparison/19863
As you can see, fine details are retained almost always perfectly to pixel level, while at first sight you can see the sky to "blend" better in the AV1 version; this is because the grain synthesis process first "scrub" the source from grain, then adds it back when decoding - the synthesis of the grain is actually encoded, but in a reduced bitrate.
At the end, it's like to have a grain plate added, but with reduced bitrate; sure, both methods are not perfect, but they serve their purpose: give an "organic" look to the image - and grain synthesis on AV1 works pretty well.
Not that anybody (but me) have a reason to produce a bitrate constrained encoding, but if one wants to put an AV1 UHD version on BD, this will surely look great at UHD-BD "low" bitrates (45/50mbps), hence perfect to fit onto a BD-50; but I strongly suspect that, with careful settings, it would be possible to put it onto a BD-25 too, up to normal feature lenghts (90/100m) of a scope film (2.35:1), retaining most of the quality of the source.
EDIT 2
I just calculated at what bitrates (avg and max) should a 100m movie be encoded, at the net of a well robust lossless audio track of 4600mbps, and the average was 27200kbps and max 42100kbps; then encoded the same trailer using these two settings, but 12 threads instead 16. PC was stuck as well, still the encoding time lowered to just a bit more than 14m - that's 6x real time, not that bad!
And guess what? The details are almost the same, as well as (synthesized) grain; you must zoom like 10/20x to spot tiniest differences at pixel level! I'm shocked!
https://screenshotcomparison.com/comparison/19877
Take this test with a grain of salt; but if the promise is kept, it will be possible to publish projects onto a BD-25 disc (without a Blu-ray compliance, of course) - or for a 32GB memory - with a transparent video result, along with a lossless audio track - or various lossy ones... if you said that 5 years ago, I laughed at you, but now...
I converted a DCP 4K trailer to AV1, using svt-av1 speed 4 dual pass... well, it took 2h27m for a 2m21s trailer - 60x!!!
Sure quality is astounding, but I can't think to let the PC convert a movie for 5 days!
Must test faster speed settings.
![[Image: AV1-Film-Grain-Synthesis.png]](https://i.postimg.cc/VvkRG5Jj/AV1-Film-Grain-Synthesis.png)
![[Image: 132940-Ozer_AV1_Figure2-ORG.jpg]](https://dzceab466r34n.cloudfront.net/Images/ArticleImages/InlineImages/132940-Ozer_AV1_Figure2-ORG.jpg)
