Hello, I have this footage:
and here is the file: Data package from June 8th. - FileTransfer.io
I have tried TopazAI, but the results are nightmarish.
Obviously, there is only so much that can be done. Can anyone suggest best settings to restore /enhance to HD (1920 x 1080) @ 23.987 fps?
The quality of that clip is rather poor.
Perhaps this might be a good starting point to create a 720p cleaned up version (which could be brought to 1080p with GAIA in a next step):
Hey, thanks!
Yes, the clip is poor…but it is the best I can get for now. What about being even more incremental. Upscale to SD and then 720 p and then 1080p?
Thats what i would do.
Source is >real< low res (with real do i mean, that it is so low, that it realy cant hold much detail in the first place. Not like with something that is 640x480 or bigger)
Shades & Gradients are there, but not rough, small texture on the skin (pores etc) are not there.
So you must create Fake Grain on the Gradients (the right amount, which is harder with new Version from Video Ai, because they use just round Numbers ! Video AI 2.6.4 can use something like 0.7 : 1.2) from which you create Fake Detail Later.
And you need to Sharpen the Edges ! without thining them Out (that’s what Dehaloing does. Another thing that it does is to shrink down the size of Pores, Pimples etc) to much.
In your screenshot i can see that you use a costum resolution.
Dont do that !
Integer Number Scaling is much better ! Compare any Output File or Preview in 1:1 Resolution (so no Zoom at all ! or if it’s toooo small to see anything, with 200% Zoom, never more !)
My first try would be something like this.
First Pass
Scale 2 x | Artemis HQ | Recover Detail 70 | Grain 1 : 1
If your output looks not Flatter, but more Sharp, and the Shades Gritty !
Then your on the right way !
If it looks just sharper, but flat (like on the right side from your Screenshot) and oversmoothed. Then Try another Recover Detail Setting, or Artemis Aliasing and Moire, or Proteus V3 (good for creating Fake Details. But try it first without Grain, to see what the Proteus Settings alone achive)
Second Pass
Now take the Output File from First Pass.
Throw it in Video AI.
Now !
Scale 1 x | Recover Detail 50 | Grain 1 : 1
Proteus V3 (not 4 !) - Manual
45, 15, 31, 6, 5, 0
Click on “add second enhancement” set Intermediate to “Scale 2x”
Set Artemis Artemis Aliasing and Moire, Recover Detail 80.
If the Output looks now Sharper, and has the same Griddy look on gradients, then your on the right way !
If not, try something different out. Until it looks sharper, but is still griddy !
Now your Clip is 960x720 and if anything was going good ^^, you can use this File,
as Source for Artemis Strong Halo V2 (with Grain 1 : 1) to make it look more HD like.
If Strong Halo thins out to much, try another Aremis Model or Proteus V3.
If both doesnt help, shrink it back to 480x360 with a Cubic Resizer, and feed Video AI again with the shrinked File, then try out some other Model.
Why shrink it down first ? What we did first, was to sharp the source, and to add grid/dirt/grain on the Flat Areas so that the perceived Detail is higher.
Now if it’s higher for us, it’s higher for Video AI too.
But just to a certain extend ! If the space between Detail is still to wide, then Video AI is sometimes not very effective !
In the Picture here you can see what i mean with Grid ! Look at her cheecks.
1 = Allready 4 times Processed (believe me you dont want to see the Source ^^) with VirtualDub2, Video AI 5 & 2.6.4 and AviSynth.
2 = Created more Grid at her cheeks ! by using Unsharp Mask in Photoshop.
3 = Played around with Gamma, Contrast, Brighntess and Colors.
4 = >Just< Shrinked Down Picture 1 with Cubic Sharp Rescaling to 60% → Same Grid Strenght as Picture 2 !!!
5 = Played around with Gamma, Contrast, Brighntess and Colors.
Grid is your friend ! Oversmoothed Areas are your enemy. Thanks to H264 Oversmoothed flat Areas are the norm !
That’s why some 8000kbps ! DVD9 from 2003 ! got way more Detail & Depth as many “4k” stuff out there that was recorded with a 1500$ IPhone and compressed & deblocked to dea**.
Details & Depth cant be without Grid !
Sharpness = Contures
Detail = Grid/Dirt !
Resolution “k” is just helping to keep Sharpness when Input must be zoomed/blown out to Display Res
(bigger Display → More k is needed, to keep the perceived Quality ! More K = More Bitrate = More GB on HD/SSD…People shoot them selves in the foot (by buying bigger Screens) and Celebrate it with “k” Trending !)
and to make the >difference between 2 Lines (that’s what sharpness is in reality) with very short distance< cleaner/better visible.
