I do quite a lot of video restoration, and as such I often have multiple copies of the same clip.
None of them are particularly good quality so I usually just pick the best of the bunch and use that as my base, then try to clean up the others and combine them with the best copy to add a bit of contextual detail and colour if possible.
I definitely get decent results from TVAI however the original source is usually in such poor condition that the end result is only that: decent. Whilst it’s certainly a big improvement, it’s not exactly good (and it certainly isn’t great).
I was thinking that perhaps something similar to Multiple-Image Super-Resolution (originally for photos) could be employed where TVAI uses multiple copies of the one clip to reinforce, sharpen and denoise the video.
An alternative to this could be to include the option to specifically train a model solely on one specific group of closely related clips.
Since all the clips I have feature the same faces and roughly the same backgrounds, perhaps the combination of all these clips into a specific model could greatly improve the quality.
Also, if there was some way to use TVAI as an add-on in Davinci Resolve, that’d be fantastic.