I’m curious, CatBus – how much do you think AI upscaling can improve print scans to get mildly closer to camera negatives quality (in terms of making despecialization more seamless)?
It depends a little on how clever the AI is. The only project I’m really following along these lines is DrDre’s, and it looks like the improvements are modest but noticeable. I’m a fan of AI enhancement, but I’m not sure it could ever really approach camera negatives.
IMO I think we’re stuck with the detail versus authenticity struggle indefinitely. Something pieced together from the best bits of HDTV, BD, and UHD will always outstrip projection print scans on detail, no matter how much we try to enhance them. But that’s the only metric they’re guaranteed to be ahead on. For despecialization, I think we’ll end up removing more detail from the high-detail sources than adding detail to the low-detail sources, to make them blend. And grain matching too, of course.
I’m firmly on the detail side of the struggle. My personal goal is to be able to watch the Star Wars trilogy as if it had had a respectful Blu-ray release (we’re talking 1080p here, I’m not picky), akin to other classic films respectfully restored for the format. Imperfect authenticity isn’t a dealbreaker for me, as long as I can’t detect it.
That said, projection print scans have factors in their favor I can’t ignore. 4K83 looks good. DrDre’s AI enhanced version looks damned good. Does it have all the detail I could ever want? No. But it’s extremely close to good enough to stop caring about it, as far as I’m concerned.
4K77? Meh, I honestly never cared for it. I don’t know much about 4K80, but that ESB print Poita had looked so good it made my teeth hurt. God, I hope that sees the light of day.