It’s probably a bit more complicated than that.

When Star Trek: The Next Generation was remastered in HD, it required a) direct access to the film reels, b) complete re-editing of the film from scratch into the episodes, and c) completely new effects, since the original ones had been done on video post-film transfer.

For a sitcom, c probably doesn’t apply, but b and even a are potentially huge costs.

If it was on film, then you had the fidelity to spare to upscale it in HD the normal way (by re-digitizing it)

Yep. It’s the reason The Original Series and The Next Generation got HD remasters, but Voyager and DS9 didn’t. The latter two were shot on digital and so it wasn’t possible, while the first two were shot on film.

But then you’d need access (and most likely licensing) to the original footage AND you’d need to find someone who even knows what ‘film’ is in the first place, with equipment and skills to use them. Not too difficult if you’re in the business I’d assume, but you can just throw it on AI in an afternoon by an intern and call it a day.

And besides now they have option to market that as ‘AI remaster’ which I suppose sounds fancy to someone wearing a suit. Who cares about consumer experience anyways.

Definitely. Assuming they kept the film and it’s in serviceable condition.

I went through a phase of testing out Topaz AI upscale tools on videos. Ultimately I didn’t like the results, as impressive as they are you always end up with some hallucinations ruining details.

The exception is cartoons. They upscale really well.

A guy did some great work with the original terminator movie. Another guy did it without using ai, but spent a hell of a lot more time on doing it. Smoothing out and fixing the stop frame motions for the skeletal terminator makes it look way better and less dated.

Yeah, most of the open source models that do the upscaling were meant for anime waifus. Also, cartoons don’t often have things out of focus in a frame like live action does.

I’ve tried upscaling an old cartoon from 1080p to 4k using some open source models and it worked quite well. However when I tried to add frames with interpolation, it added a bunch of frames that were quite horrific, especially during quick movements.

I also tried to upcale a 90s live action class so I could see some of the details the instructor was showing. It was not near as good and even without frame interpolation, there were a bunch of new artifact glitches that were added so I didn’t keep that version.

I’ve watched most of DS9 upscaled. And it was an improvement for sure. I never noticed anything strange like these examples.

But I also don’t think the upscaling/cleanup was this aggressive.

Couldn’t tell you if it was AI upscaling a normal upscaler, but the only time I’ve ever had to delete a torrent from my external hard drive was an upscaled cartoon where you could really tell it was upscaled. The outlines didn’t look right and things just looked kinda blurry and slightly fuzzy and off somehow. So, definitely depends on how it was upscaled.

Also if you’d want to upscale it anyway, why not provide source material and allow customers to use any upscaler they want?

Because Upscaling is incredibly resource hungry. You can’t do it on a 250€ “smart” TV with the calculation equivalent of a raspberry pi 2.

And then, sunk cost fallacy goes brrrrr.

I watched Daft Punk’s Interstellar 5555 AI remaster when it was in theaters recently. Granted this is an anime, not live action. I think the source footage they used was probably standard definition, but overall I’d say it was a net negative experience. There were time it would look good but when it looks bad, which is often (particularly in the background or on small objects) it’s incredibly distracting

To put it another way, the bad had a bigger impact to me than when it looked good. But then again, this is an anime and is probably pretty low frame rate especially compared to sitcoms which from what I recall probably run around 60p, so it might look better

Do you know those videos where they put a single frame of scary face as a jump scare prank? It’s very quickly flashes, so you see enough to be scared, but not enough to understand what’s happening and make sense of it. That much. It’s visible that much.

This is an interesting option that I’m surprised hasn’t made the jump from games to other media.

Admittedly they are different challenges. Games were half-res but progressive scan, video was interlaced. But hey, in the gaming world we’re at the point of adding high refresh support to emulate the CRT scan flicker. I can see a world where you create this high res 120Hz picture to simulate a shadow mask and interlaced output on modern TVs. Probably alongside a raw pixel option and an upscaled option, no reason to do just one other than storage space.

You have to start somewhere and make it better from there as you go

Here you go!