So, video upscalers, here's what I'm seeing.
There's a really good reason people are talking about video upscalers and archival footage. Archival footage is free, and it's a great restore target, because the quality is really low. A lot of it was digitized decades ago, sometimes, it was digitized from vhs or similar formats.
It's footage that wasn't designed to be shared or preserved digitally.
And so far my results with archival footage look... decent? Pretty good actually.
New problem: When you put high quality footage through an AI video upscaler, it sharpens it, and you look like you're working in a world where everything is made of rubber.
And not in a good intentional looking way either, but like peering into a universe where everything is waxy and greasy and sometimes cracked and hairy in places things shouldn't be.
My first instinct is to say, f it, and write one of my own where I can have a little more control over the granularity of the model and settings. But I think I'm going to resist that for now.
So my next experiment is using ffmpeg to degrade my video. Not to the point of total collapse where you can see the tiles, and everything looks like it's been shredded... but to a point just before that. Compress the heck out of it... and see what happens.
I'll let you know how it goes.