Hello all,
As a newbie, I would need some advices and tips to enhance some old videos.
Well, I have some Hi8/DV8 videos made in the 80’s/90’s when my children were young. And now I would like to upscale and enhance these videos. As they’re SD ones (~ 480p) I’m aware it’ll be quite impossible to turn them into 4K or more… But, at least, I would like to get HD (720p) or, if ever possible, full HD (1080p) from them. But I don’t know how to manage that…
For the full story, I made some short tests around one year ago with a trial version and I’ve been quite deceived by the results. They looked so unatural and particulary the faces looked as characters from wax museum… No details, no expressions… So, I decided to give up.
Now, few days ago I watched a couple of these videos and I felt I had to do something about them.
So, before to purchase TVAI (300 € is not nothing…), I would like to be “sure” I could get better results this time. So thats the reason why I need advices to achieve my goal. I could try your tips on a new trial version if ever possible. And if I’m happy with the results this time, then I could purchase TVAI witnout any regrets.
And if it works very well, I could try to enhance somme old VHS recorded from old TV programm as well.
Anyway, thank you in advance for any advice that can help me.
I’ve to add I tested many models but I didn’t play to much with the settings.
Hello and welcome. I recommend checking first it’s interlaced or not, Hi8/DV8 I think it is, but check this optically. Download and install Virtualdub2 if not already done. Drag the video into VDub and add filter “deinterlace”, check if it is “keep top field” or “bottom field” try both, in this sample it was “top field”
Close VDub, download and install Hybrid
Drag your video into source field, select lossless codec “ffv1”, select container “mkv”, click on destination and select output path and enter name (do not enter file extension, Hybrid does it)
open register “Filtering” and Deinterlace Tab, choose Vaporsynth QTGMC, select checkbox “overwrite input scan”, choose “topfield first” or “bottom field”, the one you find out in the prevous Virtualdub step. Select preset “very slow”, it’s very import deinterlacing happens slowly which gives the amazing results.
Go to the base Tab and start the job (when it does not start click two times on the icon or check register “jobs”)
I know this has nothing to do with TVAI, it’s pre-processing, optimal preparation of sources is very important. I recommend do every step with source files lossless. You can also import the hybrid result into Virtualdub again and there do things like resize a little bit and/or croping when needed and export it as AVI or ffv1. check here my posting Because it is SD, lossless pathway with source files is manageable, means file size gets not that huge you can’t storage it. After upscale with TVAI you can delete this process files.
I have some family VHS tapes that my father put onto DVD with a cheap digitizing machine. After doing the needed pre-processing (similar to what @Mayday described), I use Proteus Manual with these settings:
I upscale to 1920 by 1080. The enhancements are subtle, and I tend to compress them a lot for the final file, because if I don’t they come out huge but are lacking the detail to merit the file size. (For H.265 I normally use a CRF of 18, but for these I use CRF 25.)
I find taking a 480p video and re-rendering it through TVAI to another 480p video (in ProRes LT) gives me good results. I then upscale the 480p video to 720p video using other software.
For me – when it is old analog video, 480p to 720p, does not seem to provide any better results. It also takes a very long time using TVAI to do everything. I have had terrible success providing interlaced video to TVAI. Other software applications do a much better and faster job (in my opinion).
I’m wondering, when you use a reconstruction model and do no upscale, the benefit is low, additional pixels are needed a model can work. Going from 480 to 1080 is often a good scale, unless the source is so bad for models where it is better to do nothing or use a conservative model that does not reconstruct like Theia Fine Tune
Thank you all for these informations.
I’ll try the advices above sometime. I don’t have to much time at the moment to do some tests.
What I really want is to enhance some of these videos at first. Their quality is less than average considiring how are the videos nowadays. Not too much sharp, see quite blury, and a lot of grain.
If I can upscale them as well, it would be a cherry on the cake…
For the Hi8/DV8 videos, of course they’re interlaced as it was the use in the 80’s.
Having Adobe Premiere, I think I can desinterlace them with it instead using Virtualdub ?
There’s an option in this application to manage this case.
Thanks again and any more advices and tips are welcome
None of them, use VDub just for visual check if it’s top field or bottom field first. Use Hybrid QTGMC deinterlace at preset “very slow” (or placebo) as described above, it’s world best deinterlacer 
Thanks for your setting.
I quickly tried it with a short SD sequence I had under the hand and I have to admit it’s a little better than than the tests I made before. Far to be perfect but a little better anyway…
I think I’ve to play around these settings to get better results.
Here you use Proteus but what about the other models ?
Some people have reported getting good results with Iris low quality v1 (maybe v2). You can explore that if you don’t mind erroneous faces being added. I gave up trying it because it was not as good as Proteus at making the fine details look realistic or blurry. And that’s the same story with all the other models. They either overdo the fine detail enhancement so that it pops out and looks unnatural, or they don’t do anything.
I test my settings on trees and gravel or grass. Most models will make dot patterns on the grass or gravel. Trees get weird looking lines in them. Far off faces also have to be tested since some models turn them into monsters.
The Proteus settings I shared are not perfect, but a good compromise in all those areas.
I should add that if you can get those scenes to look acceptable, the rest of the video is usually fine.
Thanks for all your explanations.
I have another question (from a newbie).
Well, when you want to enhance and upscale a video (after the deinterlacing process of course), is it better to do do thes actions separetly or it does not matter ?
And, in the first case, which one first (enhancing and upscaling) ?
Thank you in advance
Upscale and enhance at the same time. Deinterlace before those. If you’re going to interpolate, do it last.
