creating larger than hd file to get better quality hd on youtube?

Given the crazy number of codecs out there, it's highly likely they're still using the FFMPEG libraries for transcoding (as is everyone else: Vimeo, FB, etc.). At MySpace we used DirectShow since the backend was Microsoft-based. Everything is transcoded for QoS (quality of service and also security reasons).

Google didn't create VP9, it's the latest iteration of the On2 codec, which Google purchased. VP10 is already out (code); doesn't appear to be in products yet. VP9 is the 'open source' alternate for H.265. However once you understand the patent system and corporations, it's all a game based on who has the most money to bend the markets to their will resulting in strange product support and crippling (I wrote two granted patents; perhaps all those years in the tech and entertainment businesses is why I'm now a philosopher :) ).

I think the reason 1080p got a recent quality boost (in Chrome and Windows Edge) was the additional transcode to VP9 instead of only H.264, where VP9 effectively provides about a 2x boost in bitrate-quality in supported browsers (i.e. not Safari lol: that's why no 4K in Safari and now likely why 1080p looks worse vs. VP9 browsers. Why doesn't Apple support VP9? See above comment about patents and corporations (does Apple support H.265? ;) ))).
 
A few folks did mention adding sharpening and noise/grain may have been in there as well. So far 1080 vp9 encoded is looking pretty good, ditto the 2048x1152 in prores hq. There are still some shots that look a little funky but most of it looks pretty good. Im looking at it full screen on my 27” imac.
 
Josh Bass said:
A few folks did mention adding sharpening and noise/grain may have been in there as well. So far 1080 vp9 encoded is looking pretty good, ditto the 2048x1152 in prores hq. There are still some shots that look a little funky but most of it looks pretty good. Im looking at it full screen on my 27” imac.
Click to expand...

josh-

glad you are seeing progress.

so what was your latest upload to yt settings for those of us on the sidelines?

thanks in advance.

rob
smalltalk productions/nyc
 
I did vp9 (mkv file type)from handbrake (from a prores hq file), cq 0 (i think thats the quality slider), medium encoder speed, highest audio bitrate. dont think i touched too much else (not in front of comp right now). 2:45 music vid came out to a little over 1 gb.

for those advocating adding noise do you mean in the compressing application or on the original exported file? how much noise? just enough to be able to see the effect if you turn it on and off, but where its not visible otherwise?
 
It does not matter what CODEC you use to upload the video for YouTube, as long as the quality is good.
Take advantage of long GOP formats because an ALL-INTRA format will simply waste bandwidth.
 
That has not been my experience. As I said the large h264, prores and prores hq files have been producing lackluster results. vp9 and the uprezzed prores hq are looking better.
 
Josh, it would be great if you could show us some uploaded examples of the different versions so we can all discuss the what we see. Maybe a short little test clip encoded with the different choices on your YT channel? More eyes might help improve for all.
 
I may. My feeling is if you have to go through it with a fine tooth comb to see differences, in a way that a typical viewer never would, then those differences are not that inportant. I was probably going to post final product on here with a mini-production diary.
 
Well, either way we all learn something. If you upload a bunch of clips with different codecs and they all look basically the same then we all do not need to fret over what we send YouTube. Just upload a short 20 second snippet encoded a bunch of different ways in a playlist and let the judging begin!
 
H.265 and VP9 should be similar in quality for bitrate. In PP CC I've been using MXF OP1a H.264 UHD (MXF) for max quality 4K uploads. From tests a while ago I found H.265 also works well for max quality 4K (and much smaller files), however encode times are much slower in PP CC, so I haven't been using it (have fast internet for uploads). I need to check bitrates/quality, however I've recently been testing QT + Cineform for proxies for the 1DX II's massive files, and after doing research verified that Cineform isn't just better than ProRes, its over 2x better in quality per bitrate: https://blog.frame.io/2017/02/13/50-intermediate-codecs-compared/ . I need to compare Cineform against MXF OP1a H.264 for 4K quality per file size as a YouTube max quality upload option.

4K vs. upscaled to 4K 1080p with sharpening (1DX II)- can you see the difference without pixel peeping?:
Note the 1DX II 1080p is fairly soft (before upscale + sharpen).

RGB 444 12-bit 1080p (C300 II) upscaled to 4K and sharpened (perhaps a bit less sharpening would be better):

YUV 422 8-bit 4K (1DX II):

For sure the last two examples used MXF OP1a UHD (the first might have as well).

How much noise to add to 1080p footage for 4K upscale & sharpen? In PP CC I'd start with around 4% (Noise Effect) and do some round trips with YouTube to see the results. This helps mostly with soft 1080p footage by providing some (missing) high frequency information before the sharpen effect, and thus gives tighter high frequency texture to increase acutance (perceived detail/sharpness, since we're not actually adding more real detail). Also be careful not to over sharpen as that makes the footage look more digital (see the above C300 II example- a little bit over sharpened, as is perhaps the 1DX II 1080p example). Best to err on the side of too soft/organic vs. too sharp/digital.
 
Thanks. One glaring thing to me is going by the screen caps (I confess I did not watch the vids...yet) is that those are CUs and MCUs of subjects. I presume theyre not moving too too much, backgrounds maybe on the static side, with the exception of that cosmic-looking background. My project is a music video with some go pro footage and a lot of handheld where “talent” or something else is moving a lot. I would assume YT compression does better with less movement is what I’m getting at. Perhaps it’s a no win situation.

I actually tried 10% “film grain” in fcpx, no sharpening, I feel it looked worse on YT.

As I mentioned there are quite a few go pro shots. I have been thinking that once a project gets exported as prores/hq etc. that anything that happened to the footage (i.e. compression) would happen to it equally, but maybe thats wrong and footage that originated in a poorer codec still suffers more? That’s been my observation with these YT tests.
 
I like using MXF for upload to YT as well. Nice to have 24bit uncompressed audio with the h.264 encoding. In fact, I have been using Panasonic's AVC-Ultra MXF format as sort of an archive format as it is 10bit 4:2:2 with 24bit PCM audio. This serves both roles as well as being a fast export out.
 
I dont have premiere cc. Do have fcpx (will see if that format is an option) and davinci but have been unable to export a stereo file from Davinci...keeps giving me separate mono tracks which youtube doesnt interpret correctly. Pilot error Im sure but I dont know where the pilots going wrong.
 
You must log in or register to reply here.
Back
Top