I reached out to various people I know from all over the internet who performed various successful upscaling projects I know of. Basically, the plan is to accumulate some additional knowledge and maybe get to know some variants in technique.
Having deleted scenes indistinguishable from at least 1080p footage would be a really big help for various fan-edit projects out there.
The results are in, mostly standard stuff like “I just run Gigapixel”. HOWEVER there was one answer in particular I want to highlight. Its from someone who is working on (and was mostly very successful) in upscaling old Full-Motion-Video Games:
For the upscaling part of the process, he is using a combination of Topaz Labs’ Video Enhance AI and Gigapixel AI. The former is specifically for upscaling video and the latter for still images.
The first step is to do a pass with Video Enhance AI, checking if there are any spots where it didn’t do a particularly good job. Afterwards, he hecks if he can improve on that sequence frame-by-frame with Gigapixel AI. That second step can be a really arduous process, obviously.
A couple of other Gigapixel-packages (DeNoise AI and Sharpen AI in particular) can be very useful - but given the resource-intensive nature of the work, coupled with the inevitable trial-and-error process to see what works and what doesn’t, it can be a very lengthy process to attempt to use all of them.
In addition, he also employs use AVIsynth+ a lot for cleaning up footage before upscaling it (since some of the original footage he is working on is terribly messing, compressed, sometimes interlaced, and so on). In particular, he mentioned to me that I might “find this necessary for some of your upscales, depending on the quality of your source footage. If they’re fully-completed-and-mastered-but-just-deleted scenes from DVD extras, they should do just fine in terms of quality - but depending on the DVD region/interlacing, you might need do a bit of that to clean up quality before you upscale.”