alt Hacker NewsI'd love to walk around with a Hi8 camera and 4K camera with a rig to record identical frames, and get a few hundred hours if footage and train a model to do intelligent upscaling and cleanup of old footage.
Possibly a 3D printed rig with a semi-transparent mirror + some crowdsourcing and this could really work.
It's in my backlog.
You won't capture identical frames to old footage. If you want to do AI slop, just do AI slop.
Old media will have flaws, errors and low resolution (both spatial and temporal). Putting an AI filter on top of that will only make it less correct.