I have been testing Video AI Models in v4.1.2 to upscale 20+ year old Sony Hi8 tape to something worth looking at on an HD TV.
I’m a non-professional underwater videographer, and my YouTube Channel has HD GoPro clips here… https://www.youtube.com/@waterdogdiver9899/videos and I’d like to be able to upload these old clips as well.
From my testing so far, the best results I’ve achieved is using the following settings…
Input 640x480 @29.97 FPS
Output 1280x720 (HD)
Frame Rate 29.97 FPS (1x Deinterlaced)
Video Type Interlaced
Field Order Auto-Detect
AI Model Dione TV
Add Noise 5
These settings will produce 5 minute video in 30 minutes that is not too horrible.
Because there are so many variables to this, I don’t want to spend weeks testing hundreds of them, and am wondering if there are good/better/best settings that others have found.
you have to forget about 1080p with such a definition, the best and fastest is to deinterlace with hybrid and qtgmc at the draft level, at the limit apply maa2 with avisynthe, export in ffv1 without loss then use proteus v4 all at zero, even the recovery of details in 1280 x 720. in any case for my 720 x 576 sources it works well, as much in 25 fps as in 50 fps. but I prefer 50 fps necessarily, it is more fluid, it hurts less to the eye and we see more details.
the sd sources digitized from old tapes do not really have compression, but more noise which is difficult to remove
Agree qtgmc deinterlace with hybrid is best deinterlace you can get, way better than doing it in Topaz. I strongly recommend this.
@waterdogdiving Then if Proteus is better or an other model, depends on source, try Proteus but also Rhea and Gaia. If you do fps changes in Topaz, additional interpolated frames are generated, yes it’s more fluid then, but if you have blury frames, the interpolated ones looks very bad! You can leave orig 29.97 when upscale, and later then you can decide with the result doing a fps increase in Topaz or not.
What I often do is bring 29.97 to 30fps with ffmpeg it’s losless, and also audio speed up 29.97->30fps audio requires a reencode, but no quality losses are audible.
can you post us the same clip, but the original Interlaced one? the one you posted you already deinterlaced it and we are not sure what AI was used. we want to try it on the unspoiled original clip.
So I made an assumption that made an ass out of me. When I captured the Sony Hi8 Camcorder tape footage to my Mac, I assumed it was interlaced. It apparently was deinterlaced by the capture device/software I used… or what I read about Hi8 tape all being interlaced was incorrect.
