"Static" noise when recording persistent color
I did now experiment with AV1 in OBs. previously I just used NVENC with a Bitrate of 60000-90000 for 4k (upscaled) 30 fps.
For my first AV1 recording I just used CQP with a value of 20, no AQ, P6 and Two Passed (qarter Resolution).
What I noticed in my footage is that in a very static screen, where nothing changes, there is quite a bit of static movement, in particular in a certain location (see circled part in screenshot, compare with video clip). I am not sure if this is considered "normal". I fiddled around with the settings, but small adjustments do almost nothing. Setting quality to 15 makes it a bit better, and AQ also makes it a bit better, and both combined are somewhat decent (file size triples). And its actually better than just setting Quality to 10. Alternatively I can of course set the bitrate to 60k-90k and its basically gone as well (with much bigger file sizes).
But I don't quite get why. Doesnt this "static flicker" even use up additional space and whatnot? Are there areny other recommendations?
33 Replies
I think the issue is there is a very subtle gradient in the color or something and it’s causing banding
Try recording in 4:2:2 or something to see if it goes away
Also the preset is basically just uselessly hogging resources when you use CQP. Use P1
Alrighty. How would I switch it to 4:2:2? I dont see any option to do so. Would I need to install like an additional encoder?
ah, found it
if it does anything to the static at all, its minimal
but generally the idea isn't bad, as the source is basically full RGB capture
but it looks perfectly fine if I do OBS fullscreen preview...
Your preview isn’t encoded
yeah, just saying that at this step its still fine. It's not like something is broken on the way into, or by the capture card itself
What capture card?
Datapath Vision RGB E1S but atm I am just capturing a HDMI signal from the Tink. That one however gets a VGA signal.
I am using a capture plugin in OBS called RGBEasy for that card
So...if I use a different process, which is using VGA->Datapath directly, then going into VCS (external program), then capturing the window of VCS, and then scaling it in OBS, then the capture is perfect. Thats how I also used it to do before the Tink
On the full screen preview both look the same basically
This is a screenshot via snipping tool of the OBS preview when doing the VCS capture where everything looks fine after recording:
(ignore the 1 cut off pixel)
thats the screen via tink which creates the issue
even if I output limited colors via the tink the recording has static
where is it?
These are both the preview pics what I see on my screen. Before the recording. As the issue only manifests in the recording.
So the top one looks fine in recording
the bottom one does not
I also tried wonky stuff now having two OBS instances and screencapturing the preview of the other ^^
(didnt help)
turn off 2-pass, lookahead and set b-frames to 0
but overall this is normal encoding things that will show up on YouTube even if you don't have it in the capture anyway
youtube's transcoding HATES minor gradients/solid blocks of color
testing, will also check with comparison uploads
Disabling this stuff makes no difference.
However drect recording does make a difference even on youtube
So here are the standard settings with CQP20
https://www.youtube.com/watch?v=rxpdqISwMwY
Here I did the changes you mentioned:
https://www.youtube.com/watch?v=toZDML3a8Do
And this here is recorded without the tink4k in between, and instead "directly" via VGA (but then screencaptured from th VCS software)
https://www.youtube.com/watch?v=4ppz17Lz5zI
I really want to use the tink though....that VCS Software is a pain to work with
Also experimented with noise reduction a bit, changed colors and not much of a benefit
then I just made screenshot of what can be seen on preview. And then I made a monitor capture of the screenshot, to exclude any movement, noise and whatnot. While its significantly better, some noise can still be seen on the recording. So...its some weird combination of things I guess...but if you have any more ideas, please let me know and I am glad to test...
For now I will head to bed though
thanks for your suggestions so far!
Still not sure what to do. But I experimented further and I think I can pinpoint it to noise.
Noise that is basically more pronounced when being encoded with a low quality.
Tried the Noise filter in retrotink (LPF), but that doesnt help. Actually might even make it worse as it feels more erratic after that.
Also tried the dithering in the retrotink. Without much of an effect.
The only solutions I have found so far are:
Setting Quality to 11. This doubles file sizes compared to CQP of 12 for my static sample recording (factor 5 compared to CQP of 20), but removes any noticable noise.
Alternatively adding adaptive Qualtization (can be CQP of 20 here). This will also make the file 5 times as big. But solves the issue as well.
Need to do a longer recording as this 1minute of a static image going through the tink is basically no movement to everything on the screen is "vibrating slightly" with these changes, just not visible to the naked eye. But I guess the factor wont we as big if there was more action going on.
But I was also wondering about a pixel noise gate in OBS...but for some weird reason that does not seem to exist.
I mean the guy who programmed the VCS software put a working version into that software. But capturing a screen of that software is dumb
Of course a pixel noise gate doesn’t exist lol
I thought you said it was not present in the preview even with the retrotink
It sounds like a difference in encoding
As someone who knows nothing about how vcs records
Seems like a program that’s only dumping AVI files
In the preview screen of obs showing the tink footage you cant see the noise. Its only becoming quite visible after recording.
But i suspect its there in the preview already and just gets magnified
and yeah...the way i recorded using vcs, was that i had the vcs pteview window on my screen and then captured the window in obs...
not elegant but it worked
Well I could not
And when I asked you said it was the preview and not there lol
Maybe a misunderstanding. This is how I would record VCS. And ofc I now artificially showed all menus and placed it all together. Usually the important thing is just that the VCS Main Window must be fully on your screen. Otherwise its not drawn and you don't get the full picture.
But just to reiterate: No issues when doing VCS recordings. Thats just my baseline. A very cumbersome program to work with because the software is not official and the inofficial one is no longer updated for PC and in fact you don't even openly get the one I am using because the developer removed it all...
And when I was mentioning the preview, I usually refer to this option:
GG @Kordanor, you just advanced to level 22 !
And there, it always looks fine. Whether I record it via tink4k or VCS.
But "looking fine" doesn't mean there is no noise. I might just be unable to see it.
All written from here onwards is aimed at recording via the Retrotink4k going into the capture card (I use Datapath as well, but just as the recorder of digital signal)
The recording then gets very visible noise (using the Tink). Unless going to quality level of 11 or using Adaptive Quantization.
That does add tons of noise in theory. But it's not perceived as noise and can only be made visible by using extreme settings in VLC Media player
Like this (from the CQP adaptive Quantization recording). You can see everything is moving around there, but only if you tweak the settings. The noise is not perceivable when watching the recording normally.
In comparison, with CQP 20 and without AQ there are much bigger blocks, and they are clearly moving around when watching the footage even without any adjustments
If I do a recording with CQP on the VCS recorded footage though, that is basically 99% clean. Even if I apply these effects to make grain visible, you can see hardly anything moving around there.
okay well
good luck
i reiterate you're chasing ghosts that no one will notice
Maybe. But I guess thats true for the majority of stuff we are doing, aiming for perfection, meanwhile people playing games in DOSBox with horrific scaling (or just scaling retroactively in OBS so that it "fits the screen"), totally washed out graphics. Or having sound crackling noboy seems to notice (Ambermoon PC Conversion is unplayable to me due to that)
That said, I just published a LP video of MM3 today, which I recorded before I noticed. And I think it's relatively visible (with timestamp):
https://youtu.be/THIyBgkJwwk?t=291
Kordanor
YouTube
Might and Magic 3: Inseln von Terra - Festung der Angst: Hebel für...
In diesem Teil: Festung der Angst: Hebel für Mumienkönig
Might and Magic ist eine Reihe von Rollenspielen von 1986 bis 2014.
Im dritten Teil der Might & Magic Reihe von 1991 erkunden wir die Inseln von Terra, auf den Spuren von Corak und Sheltem. Ein Abenteuer in einer "lighthearted" offenen Welt mit rundenbasiertem Kampf.
Playlist des Let's...
Just to somewhat to conclude this: I found a way to reduce the noise:
in the ADC Settings of the tink there is an LPF option. This in itself did somewhat help but everything but the lowest option degraded the picture overall. However in combination with increasing the decimation factor it worked.
In my case I had 320x200 and a Sample rate of 1600 and used a decimation factor of 2 previously.
I increased this to 4 now. This makes 720x400 (which looked bad prior already) unreadable, but it allows to increase the LPF (I set it to 35 now). So that in combination results in a better picture. This isn't obvious right away on the screen, but will show a significant improvement on recording.
I looked into this again after playing Stronghold (the D&D one) and the one-colored green areas of the floor there just looked terrible with the flickering.