Is there a basic starting point to apply Video AI 6 to old videos?

I have a number of videos from the early days of video recording, the mid 1980s and later to mid 90s.
They were taken with Sony and Canon prosumer equipment. They are all 640x480, frame rate 29.97. bit depth 8, and progressive. They are almost all family type shots, ie people and faces filling the frames. The res looks like these were done 40 years ago.
Is there a basic starting point of settings for improvement of such videos? The most pressing need is improving sharpness. I have played with V AI 6 and I am getting results that are very undersaturated and so much noise removal, although I don’t believe I have selected Noise reduction, that skin has lost all texture and looks very plastic. The originals look better despite their faults.
From the adverts for V AI it looks like it is a simple process, but so far it has not been that at all.
Any suggestions appreciated and TIA

Try Proteus 4, 480p to 720p, auto mode, recover details 3% or 0%. If Proteus 4 doesn’t work for your videos, try two pass upscale, first Artemis High Quality, recover details 90% or 100%, upscale to 768x576 (120%), and then upscale 576p video to 720p using Gaia High Quality. You may also try one pass upscale from 480p to 576p or 720p with Gaia High Quality if the source is not noisy. Always use FFV1 lossless output in Topaz.
Avoid 1080p, it will look worse than 720p lossless.

1 Like

Make an upscale to 720p (1280 x 720). I often try Iris first with these settings:

image

1 Like

The product team did put together a tutorial workflow video that covers going from older tape content to HD outputs over on the Video tutorials page in the Topaz Labs Help Docs Page.

Never delete your Original Files ! (Save them with a Big Size ! x264 - Preset Slower - Tune Grain, CRF 11-15) No matter how “better” the Output looks.

Gamma, Contrast & Saturation are very very Important ! (Filters like the 6 Axis Color Filter in VDub2 are very good, because with themn you can not just change Gamma, Saturation & Hue, you can change the Intensity of Red, Orange, Cyan etc !)

It should be the first Step !

If you lose Texture (Grid) keep the Resolution and try Proteus 3 (Proteus 4 has less
Risk of AI Artefacts, but agressive Smoothing)
with something like.
Grain 1:1 (will give Depth, but Compression Size will be much bigger → 2-3x the Size)
37,13,32,X,Y,0

X = If there is Noise, dont go over 8.
Y = Dehaloing will thin Out Lines, good for Hair Seperation and Contures.
But just good for Contures if you watch it stretched or upscale it.
Dehalo thins Line, Scaling makes line (by nature) wider → Lines becomes more unsharp.
Dehalo is a Pore/Texture Killer ! It’s kills Texture even faster as Denoising !

If your Output has now the same Texture Density (or better) but is sharper and Cleaner (Noise, Flicker), then try Upscaling.

Another Info:
Plastic is a result of Oversharpen+Dehalo+RevertCompression (Flat’s Out Texture/Griddy/Blocky Parts and tries to Reconstruct something that looks good. Values over 45 can be very agressive to Red Parts in the Picture. Cheecks become yellow → Flat)

If it’s just oversharpen & dehaloed & still has texture, @ Original Res and you watch at exactly 2x Size (and not strechted out to the Border from your Display)
in a External Video Player, and the Plastic Effect goes away, and it’s better as the Original File, then you should maybe keep it like that.

Always check if Proteus in Manual Mode created AI Artefacts @ Faces !
Some Setting can be good for 1 Minute, then to less or to agressive for 5 Sek.

Proteus Manual & Auto/Semi Auto use different Models, they behave different with Problems.

If i got a Source with Stable Picture Quality (so no Parts with strong blocking, just a little Bit Noise) then i try Artemis Moire first @ Original Scale and 70% Recovery, if it doesnt help, then i try Artemis HQ, and then Medium, then Low.

If i found some Artemis Setting that is good, i try Second Enhancement.
Intermediate Scale x2, Recovery 100, with different Artemis Models, and Proteus
with something like
0,10,25,0,0,0

If it doesnt look Sharper and more rich, i just do it without Second Enhancement, and accept it :slight_smile:

ps:
If Source is washed out/less Detail Density i always Activate Grain ! (this helps
especialy with Artemis. But not so much with Proteus Manual. Because Proteus Manual Guesses on 1 Frames Base)

1 Like

Thanks to all for the suggestions. Actually I am playing with the free version of Cap Cut and getting results better than with my attempts with Video Ai, but I will try the suggestions, esp not upscaling to 1080, but only to 720

One thing that caught my attention.

If the Source has strong Noise & Blocking, and u use Second Enhancement.
Then the Result that the Second Enhancement produces
(use AUTO at “Intermediate”. Value changes in “Intermediate” for Test Purposes in Preview often crash Video AI very hard)
is much better (at least with Iris & Artemis) if “Recover Detail” was >low< (0-30) in the First Enhancement, and the Second Enhancement has a “Recover Detail” value around 65.

For some strange Reason, Video AI produces more Detail & Sharpness if Recover Detail is lower (but as far as i know, AI Artefact Risk is higher)

What i like to do with SD Sources with low to strong Blocking & Mosquito Noise,
is something like this.

If it’s allready Progressive

Output Res: Unchanged !

First Enhancement:
Recover Detail: ~30
Iris Low V1 or V2 (sometimes V1 produces better Results, sometimes V2, to use
older Models, you must activate it in the Settings)
Enable Parameters->“Dynamic/Absolute”-> 0,0,0,-50,-30,0

Second Enhancement (to see the Function you must activate it in the Settings)
“Intermediate AUTO or Scale x1” (Auto normaly uses Scale x1)
Artemis HQ V12, Recover Detail 70

Scale x1/Auto will often produce a much more Detailed Output as Scale x2 (or more)

Scale x2 (or more) just produces more Sharpness at Contures, but eats up
Texture/Detail.

It is not so good visible at a still frame, but when watched with playback.