I can’t get good results, the textures of everything look very artificial, and people’s faces look deformed. I used Proteus.
What settings are you using?
I’ve found these settings to be good on several DVDs:
360 is smaller and therefore harder to enhance. There is not as much information to work with as a DVD.
If faces are deformed, try Iris model.
You may do first step via Iris resize x2, then make FHD via Proteus.
You must know that it’s impossible to make HQ FHD movie from 360p. After Topaz Video AI it may be looking good, but never like movie from Bluray.
2 Likes
I do reply in here. I think there is room for improvement to faces on low quality input. I compared diffent AI upscalers in 480p, 720p and 1080p on videos of youtube.
Got 5 Samples:
David Copperfield 480p input:
Original:
TVAI:
AVC:
HitPaw:
David Copperfield 480p input:
Original:
TVAI:
AVC:
HitPaw:
David Copperfield 480p input:
Original:
TVAI:
AVC:
HitPaw:
David Copperfield 720p input:
Original:
TVAI:
AVC:
HitPaw:
Some random video of youtube 1080p input:
Original:
TVAI:
AVC:
HitPaw:
Must say when it comes to better input quality, the faces are top notch. nothing beats TVAI but on lower quality there still need work to be done :D.
The reason you can’t get good results from 360p is that what you are trying to do is basically impossible with the current state of this tech. There’s simply no information there for the model to work with. It’s the old adage, garbage in → garbage out.
Stable Diffusion shows promise for this sort of thing in the future with nodes like SUPIR, but we are light years away from being able to do that with video and maintain temporal continuity.
2 Likes
It’s definitely possible – it’s just more difficult to do in a single pass, or using TVAI alone.
Also, size ≠ quality. Being small doesn’t automatically make something “garbage”, and being higher resolution doesn’t automatically make a video high quality.
