Youtube Upload and Streaming Quality Notes.

Decinoge · May 4, 2019, 9:12pm

This pixel art seems to look pretty well on the avc1 codec.
Even the weird effects part… still looks pretty decent.

But I guess it would work pretty well because each pixel is about the size of a brick.
That probably helps the codec not to work around so much, because codecs (from what I’ve understood) also work in blocks of pixels. I imagine the low variety of colors (blue, red, and betweens with tones to dark) also helps with that.

I have a few curiosities though:

What did you use to record that (and DS2 as well btw), and would you mind specifying the settings (more curious about bit rates, but overall would be nice)?
Is the video you uploaded straight out of the recording software, or did you edit it first? If you did, what where the settings there?

Also, when we were talking about bit rates before, I always thought the video you were working on had 60fps. But it was 30fps, so yeah… those 40MBps were for 30fps, and then you did that with 80MBps… dude… that looks amazing is what it looks like.

I’ll do a couple of tests with different grades of sharpening but all at 80MBps.
Kinda looking forward to that. I’ll post few pictures of that afterwards as well.

disestablished · May 4, 2019, 9:19pm

This laptop can’t run Dark Souls 2 at 60fps and I don’t have access to my desktop right now.
So yeah, my DkS2 videos are all based on 720p/30fps sources. And I figured out I can double the frames to trick youtube into thinking its 60fps. So I get more bitrate.

Dark Souls 2 was recorded in MSI Afterburner with the MJPEG codec set to 100% quality. Its lossy, but not bad. It compresses the color space a little bit and you can’t gamma correct it. But it runs fast on slower/older hardware and doesn’t have problems with macroblocking at 93% quality or higher.

Chrono Trigger was also recorded with MSI Afterburner. But since the graphics are so simple, I decided to use Intel Quicksync for a change. I set it to target 30Mbps. But the Chrono Trigger source files are actually 45mbps.

No editing. I just purposefully recorded for only about 30 seconds. I already knew the stairwell from Majula is challenging, for video compression. and this computer isn’t fast for processing video files. So I wanted to keep things short as possible. Short is also good, for examples.

I did not put the raw source file into youtube. For my first DkS2 comparison, I compressed the 720p MJPEG file from about 218mbps to 40mbps/720p, using Handbrake. And that 720p 40mbps file is the source I have used to make all of my DkS2 comparisons.

disestablished · May 4, 2019, 9:56pm

I actually thought I was gonna skip the upscale for Chrono Trigger. But this 720p/60 gave almost no improvement. just a teeensy little bit better color integrity. But there is actually quite a lot more improvement we could get.

Look at these pics

720p/60 via youtube

and this is the file I uploaded to youtube, to get the above pic

Decinoge · May 4, 2019, 9:56pm

Well… just a bit off-topic in here, since we mentioned DS2.
I have to say that I absolutely have the new enemy placement on the PS4/X1/PC version of SotFS.
It is more “meaningful” every once in a while (hollow soldiers hitting the giant’s trees at the forest of the giants fortress), but generally it simply made a problem of DS2 worst. More enemies at the same time is 50% harder, but 200% stupider. Also gives you more souls making the game easier on the long run.

However, and keeping in mind the 30fps which I have to admit I do not hate on the original versions of both DS1 and 2, I personally think that the DS2: SotFS for X360/PS3 (think it is also on PS3) is the best version of that game. It has the same story line, the last-last boss as in the PS4 version, but most importantly, all the enemy placement of the vanilla version.

On PC, you can’t get a version like that. SotFS is like the PS4, and apart from that you can only get the vanilla version with the all DLC does not have Aldia as the last boss.

So yeah… sorry for the off-topic but those are my 2 cent on the subject.

Decinoge · May 4, 2019, 10:16pm

I’m actually curious how that would turn out if you forced the VP09 codec with 1440p.
I guess it’s pretty much ok, but you do notice that extra “blurriness” on the screen shots.

However, I think more movement of pixels would probably screw that avc1 codec pretty good.
Even at 60fps.

disestablished · May 4, 2019, 11:00pm

2K upscale, frame doulbed to 60fps

looks way better

Decinoge · May 4, 2019, 11:24pm

Those look amazing.
Oh yeah… far from “no anti-aliasing scaling” policy, videos are another business.
An absolute necessity out here, and those (slightly blurry) pixels for a video, put no shame into the game. I take into serious account that those pictures do not stay still on the screen.

However, I decided to go wild in here.
I was being conservative with my testing and that was not the person I used to be back in college.
I have to admit that video rendering was never my thing (more of an editing guy really).
So I’ve picked up a 27 seconds section of the video previous uploaded and I decided to go by steps from 40MBps, all the way up to 320MBps.

I’m also testing Photo JPEG at 100% compression, and MPEG4 at “Best” quality. Hell, even my laptop with MPC can’t read those formats properly, but youtube eats them like cake. I also have to say about those formats that renders are insanely fast compared to H.264.

I’m also uploading an uncompressed clip @2k resolution… 16.8Gb.

So I’ll come back in a few hours and bring some screen shots with me.

disestablished · May 4, 2019, 11:37pm

I like MJPEG for gaming. Its really low performance requirement and it isn’t as sensitive to fast motion, as H.264. For those positive aspects, I can deal with some slight color space issues and maybe a small lack of sharpness.

Decinoge · May 5, 2019, 3:38am

Before the big post with a bunch of screen shots and video links, I’ll just post these 2 here.
BTW, I downed the sharp filter quite a bit. So much I would be willing to increase it a nudge.
But that will hardly be noticeable on this post.

I decided to render a Photo JPEG at max 100 quality video, and another with mpeg4, but both with 720p resolution. These did not trigger the VP09 codec from youtube’s part, but I actually wanted to test them out anyway.

720p, Photo JPEG, 100% Quality, player 1080p fullscreen scaled

720p, MPEG4, Best quality setting, player 1080p fullscreen scaled

Summary

720p, Photo JPEG, 100% Quality
https://youtu.be/BuJxab8P2Hk

720p, MPEG4, Best quality setting
https://youtu.be/jKo5ucuf_bM

I think considering all that was said ITT before, the JPEG 100 is rather acceptable for very minor/quick stuff, maybe a bit of silly sharing.
But for at least half serious archiving stuff, don’t do this at home kids.

Decinoge · May 5, 2019, 6:03am

So I did my testing, and the results were actually pretty cool.
Before anything else, the frame I captured is probably a pretty hard one for any codec. There’s a lot of detail with big contrast, both characters are moving with motion blurs all around them, and the camera itself is also moving around all that detail. Good for checking all this stuff out.

First I’ll post a sequence of screen shots of 2k videos (scaled from the 720p originals), scaled to 1080p (my fullscreen).
Afterwards I’ll post the same 2k videos, but with the 720p setting on the youtube player, and not scaled at all.
All pictures will contain the codec and the settings for the render, as well as the file size. I think those are kinda relevant to figure quality vs size (for uploads and what not).

As a note I would like to state that both Photo JPEG and MPEG4 compression won the render time by a safe margin. Those really work quick. Anyways, the screen shots:

~

1080p Scaled from 2k video screen shots

Uncompressed, RGB 8bit, 16.8Gb

H.264, 40MBps, 140MB

H.264, 60MBps, 206MB

H.264, 80MBps, 268MB

H.264, 100MBps, 332MB

H.264, 160MBps, 515MB

H.264, 200MBps, 625MB

H.264, 240MBps, 711MB

H.264, 320MBps, 790MB

MPEG4, Best Quality, 412MB

Photo JPEG, 90% Quality, 510MB

Photo JPEG, 95% Quality, 801MB

Photo JPEG, 100% Quality, 1.19GB

~

Unscaled 720p player setting on 2k video

Uncompressed, RGB 8bit, 16.8Gb

H.264, 40MBps, 140MB

H.264, 60MBps, 206MB

H.264, 80MBps, 268MB

H.264, 100MBps, 332MB

H.264, 160MBps, 515MB

H.264, 200MBps, 625MB

H.264, 240MBps, 711MB

H.264, 320MBps, 790MB

MPEG4, Best Quality, 412MB

Photo JPEG, 90% Quality, 510MB

Photo JPEG, 95% Quality, 801MB

Photo JPEG, 100% Quality, 1.19GB

~

Unscaled 720p player setting on 2k video

Uncompressed, RGB 8bit, 16.8Gb
https://youtu.be/995d7L8pWR0

H.264, 40MBps, 140MB
https://youtu.be/H69HQ17jOTU

H.264, 60MBps, 206MB
https://youtu.be/TtRKTWtLSWU

H.264, 80MBps, 268MB
https://youtu.be/5prq2G-t7m0

H.264, 100MBps, 332MB
https://youtu.be/GiS2Q-4EKaI

H.264, 160MBps, 515MB
https://youtu.be/Grbs6OYVBgM

H.264, 200MBps, 625MB
https://youtu.be/U4wOQEcHHO4

H.264, 240MBps, 711MB
https://youtu.be/XJFwNNRdQRs

H.264, 320MBps, 790MB
https://youtu.be/fmh_D3IV03w

MPEG4, Best Quality, 412MB
https://youtu.be/mPTKfgDkrYM

Photo JPEG, 90% Quality, 510MB
https://youtu.be/Bnne2bVFidM

Photo JPEG, 95% Quality, 801MB
https://youtu.be/s2BFG_J6YJo

Photo JPEG, 100% Quality, 1.19GB
https://youtu.be/4bwlqANIV0Y

~

So… conclusions.
As it is quite obvious the uncompressed version is the absolute best. Only thing better would be to record uncompressed at that resolution. If I had a powerful tower dedicated just to record… I might actually do that. But not really because only those 27 seconds of recording came out as 16.8GB… insanity. Still… it is good to have this version here just for the sake of the highest standard.

But very close to that level of quality you have JPEG 100% quality (I’ll call JPEG100).
Most of the differences are barely noticeable, a little bit of blurring on the tree at the left, right to the rock. Other stuff is even harder to detect like on the clouds, and some color changes on the sky. If you really want close to the best possible quality, but don’t want to deal with the huge file sizes, this is definitely the way to go. On this exact test the JPEG100 version was about 7% of the size of the uncompressed file.

Now for H.264. To my honest opinion, everything from 160MBps upward looks pretty damn fine.
If you are not too picky about the quality, but you still don’t want to have absolute crap… That works pretty well. Even 80MBps is pretty OK, and @disestablished did some very good tests at that exact quality.

But the difference with the 320MBps version are there, it is something you notice, specially on the actual video running. If you are aiming for best possible quality, and you have to use H.264, This would be the thing. And to be honest, setting the render to 320MBps, but in the end the original video size never goes beyond 290MBps. But that might be just on my case

But personally, I really don’t like the H.264 codec to upload. To stream and store… more than fine. But to upload to Youtube to get re-compressed? I can’t say it is the best choice out there.

However, if you compare file sizes and screen shots, for about the same size as H.264@320MBps, you have JPEG95. But the quality of JPEG95 is actually much sharper in detail. You can see details much better, and there is a general sharpness that it is noticeable in video.
However, the same does not happen between JPEG90 and H.264@160MBps. In fact there, you have a slightly better quality with H.264@160MBps. Not by much, but you can still notice it.

And that takes me to MPEG4 compared with the previous 2. The file is slightly smaller, and I bet that will be a big difference in bigger recordings. But the quality just simply doesn’t par with either JPEG90 nor H.264@160MBps.

But one big disappointment is when you set the Youtube player to 720p.
Honestly… the quality sucks D=. I took the screen shots, but it isn’t even worth watching.

So to conclude, I’ll personally use the Photo JPEG codec, either 100% or 95%.
Tending more to the 95% because the difference is very little but the file size is quite different.

With the editor that I have, and with what I have available to me, those do get me the best quality out of it. Which is pretty much what I want for the purpose of archiving these videos.

Render time is very small, and the file size is not absurd if you take into consideration that what you are looking for is quality on the final upload. Obviously, upload times are a pain, but you just leave it there for a few hours and you are done with.

disestablished · May 5, 2019, 7:25am

Wow, this is a really detailed post! Amazing.

I looked at the screenshots and video before I read your comments. Most of these are really close, but I agree: MJPEG 95% is the best setting for quality Vs. file size.

what software do you use to render?

So just to be clear: Lets say with MJPEG 95%, what codec are you using to capture? and what codec are you using for the final video to upload to youtube? Is it MJPEG 95% for both?

Decinoge · May 5, 2019, 7:34am

My capture was made with OBS, and since Sekiro is a bit hard on my laptop, I actually had to capture at 720p. That was the only way to get 60fps.
I’ve posted my capture setting a bunch of posts behind, but hey… delicious copy/pasta:

Recording Software: OBS Classic
Format: MP4
Audio Tracks: 2 (game & micro)
Encoder: NVIDIA NVENC H.264 (STUPIDLY fast, couldn’t have done 60fps any other way)
Rate Control: CQP
CQ Level: 10
Preset: Max Performance
Provile: high
Look Ahead: off
Psycho Visual Tuning: off
GPU: 0
Max B-frames: 2

So yeah, the capture is H.264.
For editing everything, because I also have some… erm… voice commentary (with a terrible microphone to do it, on a room with too much noise to do it… but it has my sofa and huge TV =D), I am using DaVinci Resolve 15. Nice free editor if you are used to Adobe After Effects, but gets some features cut off (as the sharpen effect =P).

Finally yes, to render those videos after editing the H.264 captures, I used the “Photo JPEG” codec (which I pretty much imagine is MJPEG).

So… Original H.264, Final MJPEG 95%.

Tried to capture uncompressed, but not even my SSD can keep up with writing all that data that fast (60fps).

disestablished · May 5, 2019, 7:41am

Here is 20XX, a Mega-Man X -like rogue-lite. This game is actually more difficult for compression quality, than I thought it would be! I might experiment more.

720p/30 (captured with Intel Quicksync this time, instead of MJPEG)

Upscaled to 2K/60 (I trimmed off the first part of the video. No editing software. Just started Handbrake at a later point in the video) the sound is out of sync

2 screen shots

720p/30

Upscale from 720p/30 to 2K/60

disestablished · May 5, 2019, 7:45am

I would try capturing with MJPEG and see what it looks like (I think OBS can do it). The performance is really light. You might be able to do 1080p.

Decinoge · May 5, 2019, 7:52am

Nop. At least my OBS doesn’t (but I had SLOBS for a while and I can’t remember it having).
Only 2 NVENC options for H.264, and software H.264.
And yeah, for recording… which is kinda weird.

But the thing is that Sekiro is in fact a bit heavy on your PC. Not hard to get it to the top, and still get the 1080p and 60fps while playing, but recording? I had to lower everything to medium settings, and 720p resolution.

I also recorded a Dark Souls Remaster run. And I had no problems playing DSR at 1080p and 60fps, recording at 1080p and 60fps, and look amazing.

disestablished · May 5, 2019, 7:56am

Huh, that’s strange. Usually Quicksync and MJPEG are only a few FPS hit on performance. Although 60fps recording is more demanding.
I’ve never used OBS before, but it sounds like its a hog. Are you only recording or is this streaming, too?

What about Nvidia Shadow Play for recording?

The smoothest recording program I have ever used, is Bandicam. Its not free, but its really good for recording. Its very efficient for performance. *I think you can try it for free, for testing. It puts a logo on your video, for the free testing.

Decinoge · May 5, 2019, 8:05am

Yeah I went around checking stuff to get better recordings without hitting my performance too bad.
Shadow Play… I just couldn’t find how to give specific setting for how I wanted to record beyond some very simplistic sliders on a terrible “overscreen” GUI… so I quited on that one after checking the recording kinda sucked.

I did tried Bandicam… but yeah… big logo. I’ve been making an effort to go “clean” lately, so I didn’t looked for a hacked version. Just went pass by it.
Also read about a handful of other options… but almost every single one was paid.

OBS seems to be way to go for free.
But like I said… DSR really did well, all with 1080p@60fps, so I mostly complain on Sekiro than anything else.

Kinda curious to know how OBS will behave with DS2 and DS3.
But still, if I could, I would take something that records MJPEG, at least at that high quality. But I can’t be sure it will record properly.

Also the thing about OBS with NVidia NVENC is that it is insanely fast. But still H.264.

disestablished · May 5, 2019, 8:16am

MSI Afterburner is free. Can be a little annoying to use, but it has tons of features and pretty good options for recording.

Including the ability to pre-record to RAM. What that means is, you can allocate RAM space as a temporary buffer. Its useful if you don’t want to save everything to your harddrive, but still want to record the highlights.

So, lets say you know that your recordings will take about 1gib per minute of recording: you could set Afterburner to constantly buffer video capture to 2 gigs of RAM. Then if you had a cool kill in an FPS or something, you just hit the capture button and the last 2 minutes are instantly saved to your hardrive-----and it continues to record to your hardrive, until you press the button to buffer, again.

*A tip with Afterburner, is to set it to cap your framerate while you record. If you don’t, sometimes the video isn’t smooth. Especially if your computer is already maxed out on the game.

Decinoge · May 5, 2019, 10:47pm

The upload finally finished, 40GB for a 26:30 video… kinda exagerated but whatever.
Quality is MJPEG 95, PCM audio.

This video has comments, and I would actually be interested in getting some opinions on it.
Should I keep my commentary or not?
Should I try (for a few more days, been on it for days already) to make the voice clearer (very very hard ;___;, no matter what I do I lose a lot of data because of noise, and I don’t want the “voice on a bottle” effect)?
The quality is good, bad, okish?

I increased just a tiny amount on the sharpness of colors from the previous test videos (really, only one value out of 50), and I think it does at least look pretty nice right now.

I also have a non commentary video, which has the exact same length… but now that I think about if I go no commentary I’ll at least do some further cutting around the video.

Again, I’m not trying to be a youtuber in here. My primary goal is archive, in which sense my own reactions while playing the game seem rather meaningful. I also thought that I can have both commented and non-commented versions uploaded in there. But that sounds to me more and more like a half assed measure, and I should simply compromise with one way or the other.

Also a second goal with this recording is to actually share with you guys, and who else might be interested, how I “feel” these games. I am never as good as I wish with words, but I think the way I play a game is a much more straight forward expression on what the game is to me.

Anyways, I don’t want to bother you guys, so don’t check if it is too much of a hassle.
But I will be thankful if you do watch and give your opinions on the subject, and I’ll try to learn as much as I can from what you say.

Commentary
Video Deleted

Non Commentary
Video Deleted

[edit]
Getting a better microphone is not exactly an option right now.
If I could spend money on that kind of stuff, I would first get a capture card (some Elgato equivalent) so I could record some games directly out of hardware, like PS3 and X360.
[/edit]

[edit2]
Videos are off because I’ve decided to go full ballzie and just upload with comments.
Kinda made my way with the sound editing on the voice, so I’ll just risk it
[/edit]

Decinoge · May 6, 2019, 12:11am

Yeah, almost completely forgot to answer you.
Got the program, but still haven’t run any test.
just can’t get it to actually… record anything Oo.
I’ll dig further into it, but I was very happy with the big array of codec choices.

I can’t get Quicksync though.
My laptop doesn’t even have a VGA output, and no motherboard embedded Intel graphics card… so… no Quicksync.

It can however also use NVENC which after benchmarking it (lovely feature for external plugin codecs) can record 319frames @1080p per second. That is rather cool, but obviously, without any game running wild on the graphs card.

I am kinda sad that I can define the recorded audio as PCM, I actually have no idea what format the sounds gets recorded because I still haven’t been able to record anything.

But since my next recording (if I get to it) will be DS2, so I’ll do at least a few tests in Majula, using the MJPEG codec.