I’m not gonna lie, I edited the h264_vaapi encoder on kdenlive to use hevc_vaapi instead, and benchmarked that on yt, and noticed that the quality on 720p was somehow better than the native (downscaled through OBS) resolution, contrary to my previous tests, I’m not sure why.
1080p looks borderline identical to 720p though but that was kinda my intention, it’s a neuron activation option with a very slight increase in quality if people wanna use it, that doesn’t compromise the other qualities.
Recording on OBS with x264_vaapi though? Yes that did look absolutely horrendous during my testings.
That is interesting, unfortunately as I’ve already mentioned, I don’t really have the time to render the video twice. I need to leave videos rendering overnight to upload when I wake up, there’s no timeframe where I can afford to render all the jpegs back together (or the video with cas).
Please be clear that if you do this, it’s 100% for click trolling Quality Snobs. There can be no “slight increase” in quality - at very best you might limit the guaranteed decrease to be “only slight”.
There are many algorithms for upscaling images, and some are much better than others, but none of them can improve the quality of the original master. All they can do is make the distortion artifacts introduced by upscaling appear visually less repulsive. They cannot recover detail you lost in previous downscaling or lossy to lossy transcoding operations. All the best of them can do is create an additional loss of fidelity that is prettier than what a more naive algorithm might add.
It’s not rocket science any more to create an algorithm that can repaint a digital image of a child’s crayon drawing in the style of Da Vinci - but it hasn’t “added quality” to the fidelity of the original digital image… If that’s what you want, that’s ok, but that’s digital art, not high fidelity.
I have no idea what you’re talking about, all I do is set scale=1.5 with some hardware accel settings on kdenlive, you’re talking like I’m generating pixar posters on bing like a child, lmao.
Yes I understand that it doesn’t make sense that upscaling can “recover” lost quality from the downscaling I do when recording, however I also did a lot of testing and noticed that the image looked better on 720p (and slightly better on 1080p) when I upscaled the videos instead of just rendering them on 720p.
I still need to do more testings to pinpoint how, though maybe someone who knows how the codecs work may know why.
People who know how codecs work won’t be able to help you with that - you’ll want the “CD’s can never compete with vinyl records and valve amplifiers for quality” crowd to make that argument.
You may find the upscaling artifacts pleasing, but you haven’t created higher quality source material, just higher bandwidth usage for something that could be done just as well or better at the recipient’s end …
you have to look at “Real-ESRGAN” algorithms ( AnimeVideo-v3 preset can easly increase resolution of all cartoons and similar ) … nothing to do with video editing software.
best way i have found to have them precompiled (and usable with an RTX4060) was to use Hybrid project (made by Selur) on Windows
As I’ve said a couple times already, I’m not going to use AI upscaling. Even if I wasn’t anti-AI, if I had a machine powerful enough to run these algorithms on 25 minute videos (or 50 minute raw recordings), I would just, record at a native 1080p resolution, or even 1440p at that point.