>>8169628
Did some more testing, even if I use 2 models like Ani4K and then ESRGAN, it's still more than twice as fast compared to using ESRGAN normally and the results appear to be cleaner.
Some samples: https://slow.pics/c/VpyrvD0x
Especially in dark scenes like sample 3 it's very clear that Ani4K+ESRGAN does a better job than ESRGAN by itself.
Either way, I need to do more testing with other shows.
Bible Black will get a big upgrade anyways due to downscaling to 480p rather than 720p, since the source I used was a shitty "remaster" which was just an upscale. I should probably have used the original.
>>8169915
Just the Mega, so spreadsheet and links to torrents are down, but the upscales are still on Telegram.
>>8170320
Lanczos is not an upscaling algorithm, regular scaling algorithms pale in comparison to that.
Lanczos like other ordinary scaling algorithms will just make whatever blur is in the original video bigger, aswell as all the artifacts and the noise.
Here's a sample comparing an upscale to the original video: https://imgsli.com/MTM2ODEw
I don't know what scaling algorithm imgsli uses, but there's not gonna be any significant difference of lanczos to other shit like bicubic, as long as it's not using nearest neighbour, which I don't believe it is.
A few more samples:
https://imgsli.com/MTM2ODE0
https://imgsli.com/MTM2ODIx
https://imgsli.com/MTM2ODI0
https://imgsli.com/MTM2ODI3
https://imgsli.com/MTM2ODM0
https://imgsli.com/MTM2ODQx
Ignore the color change, it's samples from before I fixed them.
Lanczos will be more faithful to the original as it's just stretching the existing pixels. It will not increase clarity, fix noise, reduce compression artifacts or eliminate banding.