So you upscaled the frames from the result video and not the faceset pictures ? If so Remini is clearly the best result but what a nightmare it would be to upscale 15000 frames or more for a 10-20mn video. However it's doable for low quality pictures of a faceset, as it's usually a few hundreds/thousands of pics. But for a long video, DFDNet seems to be the only practical solution, even if doesn't look as good.
Btw, I would be interesed to see a comparative vid made with the different AI engines in TOpaz video enhance AI. They are also practical to use, just like DFDNET. In fact it's even more simple. there's a GUI and all, but not sure if DFDNet is still superior for faces.
Topaz results are worse then DFDNet, especially side pictures and usually problem with the teeths, best results come out DFDNet first, then Topaz for enhancement and upscale if necessary, the combination of using both seems better results then using them solo
I have been doing that for months and I agree that using DFD -> Topaz seems to be best. However sometimes using Topaz may be overkill since it may make the faces look too plastic.
Is there a practical way to upscale the merged faces and somehow stick them back on to the video? I've been using DFD for src face only.
PornSites.xxx
PornX
DeepSwap
If so Remini is clearly the best result but what a nightmare it would be to upscale 15000 frames or more for a 10-20mn video.
However it's doable for low quality pictures of a faceset, as it's usually a few hundreds/thousands of pics.
But for a long video, DFDNet seems to be the only practical solution, even if doesn't look as good.
Btw, I would be interesed to see a comparative vid made with the different AI engines in TOpaz video enhance AI. They are also practical to use, just like DFDNET. In fact it's even more simple. there's a GUI and all, but not sure if DFDNet is still superior for faces.
However sometimes using Topaz may be overkill since it may make the faces look too plastic.
Is there a practical way to upscale the merged faces and somehow stick them back on to the video?
I've been using DFD for src face only.