If you're looking to get software to batch process a bunch of old video through, you could do that, but I think you'd be really disappointed. Someone spending multitudes of hours (using very expensive computers) remastering a single video could produce results where you'd say, wow, that looks pretty good. But otherwise, no.

Interesting question though. And it is sad that so much old porn is not cared for. When upscaling technology does get insanely good, there will be just a small fraction of porn source material to upscale.

[Link]

picture upscaling doesn't work as in the movies (Enhance - Enhance - Enhance). There's only a limited amount one can do, simply because there isn't more information other than guesswork.

video upscaling on the other hand has a major advantage: you can use the surrounding frames to enhance each frame. So from 10 blurry images that show similar, but not 100% identical stuff, you can make 10 sharper images (and so on).

You can think about it this way: in one image you see a blurry black spot. besides a sharp white circle. In the next image you see a sharp black square and a blurry white spot. Assuming those two images are from one and the same object, you can now make two enhanced frames, and you know there is a black square and a white circle on both of them.

[Link]

The scaling quality is only as good as the video quality you're trying to upscale though.

[Link]

SVP (easily found through Google) is great but takes a lot of processing power, or time if you're rendering a new file, but the results are very good.

[Link]

Playing devil's advocate, what's to say you didn't misunderstand or misinterpret the examples you claim exist and would thus be sending adtlogin on a wild goose chase? Your preface of not being an engineer is surely an admission that you don't intimately understand the way the upscaling works so how can you claim to know the extent of its capabilities? You just cite the AI buzzword and say it's "cool" then use rhetoric of time and resource being dedicated to upscaling (history has shown many times that doesn't guarantee a practical result e.g. the Space Shuttle) but you haven't given any explanation how you make more from much less with such a big gap in resolutions such as the 360p to 4K example given. By providing the links to the stories you said you like to keep up on, we'd be on the same page assuming you're correct.

I could claim I wrote a codec that can make a 2 hour 4K blu-ray video fit on a 1GB flash drive and decompress with pin-sharp clarify. Should I not be expected to prove that or does the burden of proof lie with the doubter?

[Link]

I'll explain AI upscaling and then provide a couple of links below. A lot of the explanations I found online were convoluted. The "AI" really refers to "machine learning", whereby a computer detects minute patterns in the data it's given. In this case, the machine is given 1000s upon 1000s of high-res images. This is a porn site, so let's use tits as an example.

The machine observes tits of all shapes and sizes, from different angles, and on a variety of women...different ages, proportions, you get the idea. It could tell you the arrangement of pixels on someone's left nipple. Then you take all those 1000s upon 1000s of images and you downscale them to tube TV quality. The computer will now map those low-res images to the high-res ones. Boom, done. Now when you feed the computer a brand new low-res image, it'll say, hey---those tits looks a lot like the ones on that 25 year old Hungarian beauty. I'll use that data to fill in the missing pixels and turn this one into a high-res image. That's all there is to it. I'm simplifying, but you get that. What the computer is analyzing is a lot more abstract than tits. In any case, the results are amazingly good.

There are hurdles. The first hurdle is video. The computer can upscale every frame of a video, but the problem is that it's going to upscale each frame slightly differently, like each frame is a fresh picture. There are ways to solve for that, but video introduces a new dimension. You want to upscale the image, but you also want to adjust it so that it appropriately matches the previous frame. All of this is done. It's already here, it's already technology. There are lots of pictures to give to the computer to analyze. The computers that do this work are expensive (but getting cheaper and better each year), plus you need the programmers and hardware engineers. But there's also lots of money to be made in upscaling low-res videos. You've got the archiving and history people, the survelliance and security people, the entertainment biz, and also TV manufacturers. The latest 8K TVs already do this upscaling in a more rudimentary way. 360p upscaling for everyone isn't 20 years away, it's more like less than 10...and potentially way less than that if you're into the latest gadgets and have money to blow.

Some links:

Enhancenet

General article on AI upscaling

8K TV AI upscaling (consumer type article)

Edited by - zingerbrook on 5/10/2019 8:38:32 PM

[Link]

Nvidia's AI Makes amazing slow-mo videos

Edited by - leatherface on 5/24/2019 8:39:09 AM

[Link]

Edit: Purchased at Best Buy. I'm going to that thread about defunct sites we miss, join one that's still operational, and see how how those videos look. I've read that the AI upscaling on the NVidia is best for digitally-shot media. We'll see...

Edited by - zingerbrook on 8/26/2021 11:16:48 PM

Edited by - zingerbrook on 8/26/2021 11:17:18 PM

[Link]