Wanted to show you guys a comparison with the AI Enlargement. First I want to temper expectations for what this tech can currently do - medium to closeup shots of people can benefit greatly - especially if the image takes up the full 480i image. Wide shots and 480i Letterboxed sources still benefit, but the neural network enlarger has less to work with.
Also - before feeding the video to the AI enlarger, I had already performed countless hours of work cleaning and selecting the correct frames from my recording of the sped-up TBS broadcast. Then I aligned the images of my recording and the VHStoDVDR copy, sewed the expanded left side and tried my best to merge any fine edge detail to the image of my recording. Then there’s several different ways to feed the footage to the AI enlarger and it took several run-throughs before I got the thing latching onto the available detail and then doing something with it. It’s definitely not something you can just feed any type of VHS footage into at any resolution. Then it comes out looking too unnaturally, artificially sharp and it requires more processing in Avisynth. So with that out of the way: