I've been re-watching star trek voyager recently, and I've heard when filming, they didn't clear the wide angle of filming equipment, so it's not as simple as just going back to the original film. With the advancement of AI, is it only a matter of time until older programs like this are released with more updated formats?
And if yes, do you think AI could also upgrade to 4K.
So theoretically you could change a SD 4:3 program and make it 4k 16:9.
I'd imagine it would be easier for the early episodes of Futurama for example due to it being a cartoon and therefore less detailed.
I think it would be possible. But adding previously unseen stuff would be changing/redirecting the movie/show.
Each scene is set up and framed deliberately by the director, should AI just change that? It's a similar problem like with pan-and-scan, where content was removed to fit 4:3.
You wouldn't want to add content to the left and right of the Mona Lisa, would you? And if so what? Continuing the landscape, which adds just more uninteresting parts? Now she is in a vast space, and you already changed the tone of the painting. Or would you add other people? This removes the focus from her, which is even worse. Well this is just a one frame example, there are even more problems with moving pictures.
It would be an interesting experiment, but imo it wouldn't improve the quality of the medium, in contrary.
But adding previously unseen stuff would be changing/redirecting the movie/show.
You could see this with The Wire 16:9 remake. They rescanned the original negatives that were shot in 16:9 but framed and cropped to 4:3. As a result the framing felt a bit off and the whole thing felt a bit awkward / amateurish.
Sometimes thats true, but not all things in a shot are very important. There may be buildings or plants or people whose placement in the shot is not important. They only exist in the shot to communicate that the film is happening in a real living world. 99% of directors don't care about where a tree in the background is, unless the tree is the subject of the shot.
Ai improving a shot would be debatable, but it is definitely possible. 4:3 media on a 16:9 display is pretty annoying to most people seeing the black bars on the sides. Even if the AI only adds backgrounds or landscapes, simply removing the black bars would be an improvement enough for most viewers.
Holy cow that is beyond impressive. Sure enough, sometimes it does hallucinate a bit, but it's already quite wild. Can't help but wonder where we'll be in the next 5-10 years.
Eh, doing this on cherrypicked stationary scenes and then cherrypicking the results isn't that impressive. I'll be REALLY impressed when AI can extrapolate someone walking into frame.
The video seems a bit misleading in this context. It looks fine for what it is, but I don't think they have accomplished what OP is describing. They've cherrypicked some still shots, used AI to add to the top and bottom of individual frames, and then gave the shot a slight zoom to create the illusion of motion.
I don't think the person who made the content was trying to be disingenuous, just pointing out that we're still a long ways from convincingly filling in missing data like this for videos where the AI has to understand things like camera moves and object permanence. Still cool, though.
A proper working implementation for the general case is still far ahead and it would be much complex than this experiment. Not only it will need the usual frame-to-frame temporal coherence, but it will probably need to take into account info from potentially any frame in the whole video in order to be consistent with different camera angles of the same place.
You should be able to but remember that aspect ratios and framing are done intentionally so what is generated won't be at all true to what should be in scene once the frame is there. You'd be watching derivative media. Upscaling should be perfectly doable but eventually details will be generated that will not have originally existed in scenes as well.
Probably would be fun eventually to try the conversion and see what differences you get.
I was just thinking that. Or something like a comedy bit where the camera pans to a character who had just been out of frame.
Overall it seems like impressive technology to be able to reform old media, but I’d rather put it to use in tastefully sharpening image quality rather than reframing images.
Exactly, and to add to it, you can't know the director's vision or opinion on how the framing should be adjusted. AI can make images easily but it won't understand subtext and context that was intended. No time soon at least.
Very true, I remember a few years ago someone converting old cartoons to a consistent 60 frames a second.
If they’d asked an animator they’d have found out that animation purposely uses different rates of change to give a different feel to scenes. So the improvement actually ruined what they were trying to improve.
Yes, sometimes frame rates are intentional choices for artistic reasons and sometimes they are economic choices that animators work around.
Old Looney Tunes used a lot of smear frames in order to speed up production. They were 24 frames per second broadcast on doubles, which meant 12 drawn frames per second with each frame being shown twice. The smear frames gave the impression of faster movement. Enhancing the frame rate on those would almost certainly make them look weird.
If you want to see an artistic choice, the first Spiderverse movie is an easy example. It’s on doubles (or approximates being on doubles in CG) for most scenes to gives them a sort of almost stop motion look, and then goes into singles for action scenes to make them feel smoother.
Definitely. I remember hearing about The Wire being released in a 16:9 format, even though it was shot and broadcast in 4:3, and how that potentially messes up some of the shot framing.
They did I by cropping from top and bottom rather than AI infilling, but the issue is the same.
IIRC, David Simon wrote a really interesting piece about how they did it but did everything they could to try and stay true to Robert Colesberry's carefully planned framing, as they were aware that had it been intended for 16:9 he'd have framed things differently.
Personally I wish they had kept it at 4:3 and only released it in a higher resolution. Glad I still have my old 4:3 DVDs.
Previous century I bought a TV that was 16:9 and had software to stretch 4:3 broadcasts to fit the screen. It chopped off a tad at the top and bottom and stretched the sides a tad to fill the screen... it was horrible. I'd rather have the dark borders at the sides over mutilated images. Somehow I doubt AI would be a bit more creative.
Yeah it's always better to watch the original aspect ratio. I remember renting VHS movies and being frustrated either way because the screen was so small. That isn't a problem these days.
Why would you want that? It's always best to consume media in its purest form, and that means with its original aspect ratio. Resolution is something I'm flexible on, because I figure that filmmakers and tv directors in prior eras would have used HD if it was available, but aspect in general is tied to format, though it can be used to great effect to convey the space of a scene in different ways. Changing the ratio is akin to changing the color pallet. Might as well offer Instagram-style filters for older content while you're at it.
Because adding more detail to the sides doesn't change the quality of the show? If anything, it improves it.
I'm also not going to install some old CRT monitors to "consume media in its purest form", because I want the best quality, not the quality the filmmaker wanted back in the day it was created.
The best quality IS the quality the filmmaker wanted when it was created.
People buy CRT TVs even now btw, the absence of input lag makes it ideal for competitive smash bros.
Most people are perfectly happy watching an old film/show on a new screen because it can near as dammit replicate exactly what the director wanted us to see. Adding actual stuff to the framing of shots is a nuts idea and it absolutely changes the quality.
To give an example, having someone's face fill the screen versus being centred with more background visible around it changes the feel of the shot a LOT. 12 angry men is a good example here because the camera gets almost imperceptibly closer to the person speaking throughout the film to ramp up tension and draw the viewer into the story. It creates a feeling of claustrophobia that would be dulled if you added content to the edges of the screen.
Who is the person that enjoys old shows, but also can’t get past the old aspect ratio?
If the AI is just adding complimentary, unobtrusive parts to the shot, so as not to disrupt the original intent, I have to ask- is there really value being added? Why do this at all?
George Lucas thought CGI could make the original Star Wars movies better.
A similar thought I've had is AI removal of laugh tracks (maybe introduce background based on non-laugh track scenes)
This would make old Scooby doos actually watchable for me, so I can judge modifying the original a little if it'd what someone prefers - you can always just not watch it
Thinking about it a bit more I 100% would be the type to use a 16:9 ratio, just cause I hate black bars
I’m not 100% against tweaking old media (so long as it’s an alternative to the original rather than a replacement), I just think the effort and outcome of widening shots is misguided. More for live action than cartoons. Like if an actor is entering a scene from a side are we going to trust the AI to perfectly add them and merge them without it looking obvious and weird?
For removing laugh tracks, the Scooby cartoons might be a good case for this. They seem paced properly without them. A lot of sitcoms pause for laugher so removing it makes the pacing weird. I’m sure the Big Bang Theory video with laugher removed proves how nightmarish scenes can become (you know in addition to being Big Bang Theory).
They were shot on negative 35mm film, and edited on tape in that aspect ratio.
Only some odd scenes and shots were captured on tape, but it was mostly film. They won't need to upscale any live scenes, but they would have to work on rendering all the digital effects and blue screen shots for example. As it was only ever edited on tape, it's unlikely that the digital effects could be ever rendered or upscaled, it would probably need to be entirely reworked from scratch.
Very daunting task.
Although, they did it for TNG. ¯\_(ツ)_/¯
Edit: Someone pointed out the obvious in a reply. Yes, of course they will need to be scanned for a better resolution scan. I already pointed out it's on film, and even mentioned why it's not just that simple. (Hence the digital effects needing to be redone).
It's worth noting that edited on tape also means that to make hd versions, you have to re-scan all the camera negatives (if they even still exist), then re-edit all the scenes to be exactly the same editing as the tape editing. Plus, all the colour timing needs to be redone.
It's a huge amount of work, and the tng stuff wasn't profitable because of it. It's just too much.
Yeah, but why would you want to? It would have to generate new imagery to fill out the gaps. That's bound to not look right. It at the very least would not be fitting the artist's intention.
Adding imagery that reliably looks good is currently beyond what AI can do, but it's likely going to become possible eventually. It's fiction, so the AI making stuff up isn't a problem.
Upscaling is already something AI can do extremely well (again, if you're ok with hallucinations).
I'm not sure it's really beyond the scope of AI. Stuff like stable diffusion in-painting / out-painting and some of the stuff Adobe was showing off at their recent keynote shows were already there.
Those are on a completely different level from having someone walk into frame though, and they still only work on small things that can be extrapolated from the image.
I used adobe's Generative Fill tool (where it uses AI to fill in the blanks like adding more sky/backgrounds/hiding people) and it's pretty miss. 20% of the time, it would kinda work. And I say that loosely.
I think in a few years, we'll get there. But not today.
its already possible, but not really good atm. saw some scenes, where they changed the aspect ratio from 16:9 to tiktok. im waiting for realfill, which google is working on. idk how it works, but it can guess, what should be outside of the image and the images i saw were pretty amazing compared to SD. 4k upscaling at 120hz also no problem, as i did this to the first season of spongebob and it really works fine with my hobby equipment in one night. id really love to see an upscaled 16:9 version of the it crowd and the first 6 seasons of scrubs.
For Buffy they recut the shot often using the raw footage, and they did so very cheaply so film equipment was often visible. They also didn't address how bad the make up looked in HD, but then soft focus face filters are also garbage.
The Simpsons when they tried they cut the frame, which is just laughably bad as it removed information and the context for scenes.
When the Wire was done they got David Simon back to work on the conversion, he considers it a completely different cut of the show. I think this is the only way to do it as it means re framing the shot, this is a decision for the director, editor, and DP IMO
AI making shit up to add to the frame removes the context for the shot. Nothing wrong with black bars for me, I just want good colour balance and upscaling.
It works well enough to be shown for a few seconds at a keynote for static pictures in some cases. It won't yet work well enough to be permanently known as the official "remastered version" for moving video consistently.
Now, someone uploading a watchable version on YouTube? That will happen in the next years if it hasn't already. But that version would be widely ridiculed if released officially because something, somewhere will be off and fans will notice.
Isn't that the real beauty of AI? We can go over the parts that fans flag as distracting or obviously wrong, and release a new version. This would have to be done on some streaming service so no physical copies until a finalised version everyone is happy with.
I guess it could be done as a Bandersnatch-style experiment on one streaming platform, but it's so far from the regular way things are done that it seems unlikely.
I don't understand why there are so many purists bashing the idea. To answer directly, it can be done and it can be done well enough, albeit you'd need a fair amount of GPU. There's a reason it's mostly used for still images. Also, don't expect scene to scene consistency. New hallways, etc, unless we get models that can account for that.
Also yes, you can upres stuff, even before all this image stuff. It's gotten pretty good, even for video-- usually benefits from some human touch ups though; it's used quite well on some video games. That's easier than out painting, but has trouble with text and finer designs that are harder to guess.
Plus, it's arguably easier with live action rather than animation although I'm not certain. Redoing an art style isn't always easier than just filling in scenes.
Honestly at the rate we're going we could probably even have AI generate whole episodes of stuff, but that's when we're really going to be treading dangerous waters.
I have an old Panasonic plasma and I was always a bit worried when I watch Star Trek (right from TOS) that those black bars on the side of the screen will damage the panel.
I have 3 old Panasonic plasmas, they are the greatest tv's ever. My wife wants an led smart tv, she will have to pry these tv's from my cold dead hands. They weigh so much she can't move them herself