Do you mean something like this? (warning: reddit link)
this post was submitted on 29 Oct 2023
100 points (87.3% liked)
No Stupid Questions
35703 readers
4109 users here now
No such thing. Ask away!
!nostupidquestions is a community dedicated to being helpful and answering each others' questions on various topics.
The rules for posting and commenting, besides the rules defined here for lemmy.world, are as follows:
Rules (interactive)
Rule 1- All posts must be legitimate questions. All post titles must include a question.
All posts must be legitimate questions, and all post titles must include a question. Questions that are joke or trolling questions, memes, song lyrics as title, etc. are not allowed here. See Rule 6 for all exceptions.
Rule 2- Your question subject cannot be illegal or NSFW material.
Your question subject cannot be illegal or NSFW material. You will be warned first, banned second.
Rule 3- Do not seek mental, medical and professional help here.
Do not seek mental, medical and professional help here. Breaking this rule will not get you or your post removed, but it will put you at risk, and possibly in danger.
Rule 4- No self promotion or upvote-farming of any kind.
That's it.
Rule 5- No baiting or sealioning or promoting an agenda.
Questions which, instead of being of an innocuous nature, are specifically intended (based on reports and in the opinion of our crack moderation team) to bait users into ideological wars on charged political topics will be removed and the authors warned - or banned - depending on severity.
Rule 6- Regarding META posts and joke questions.
Provided it is about the community itself, you may post non-question posts using the [META] tag on your post title.
On fridays, you are allowed to post meme and troll questions, on the condition that it's in text format only, and conforms with our other rules. These posts MUST include the [NSQ Friday] tag in their title.
If you post a serious question on friday and are looking only for legitimate answers, then please include the [Serious] tag on your post. Irrelevant replies will then be removed by moderators.
Rule 7- You can't intentionally annoy, mock, or harass other members.
If you intentionally annoy, mock, harass, or discriminate against any individual member, you will be removed.
Likewise, if you are a member, sympathiser or a resemblant of a movement that is known to largely hate, mock, discriminate against, and/or want to take lives of a group of people, and you were provably vocal about your hate, then you will be banned on sight.
Rule 8- All comments should try to stay relevant to their parent content.
Rule 9- Reposts from other platforms are not allowed.
Let everyone have their own content.
Rule 10- Majority of bots aren't allowed to participate here.
Credits
Our breathtaking icon was bestowed upon us by @Cevilia!
The greatest banner of all time: by @TheOneWithTheHair!
founded 1 year ago
MODERATORS
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.
Great points. I agree.
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.
It is the first iteration of this technology, things will only improve the more we use it.
That it can do still images is already infinitely more impressive than not being able to do it at all.
just fyi, your link is broken for me
i wonder if it's a new url scheme, as i've never seen duplicates in a reddit url before, and if i switch it out for comments it works fine
Thanks! Fixed
i wonder if it’s a new url scheme, as i’ve never seen duplicates in a reddit url before
I think you're right. It should work with the old frontend (which I have configured as the default when I'm logged in):
https://old.reddit.com/r/StableDiffusion/duplicates/14xojmf/using_ai_to_fill_the_scenes_vertically/
that's weird. it's actually a pretty useful feature, but it's odd they'd add it to old reddit before new reddit, considering it's basically deprecated. maybe it's just an a/b rollout and i don't have it yet
i have old.reddit as default as well, but i'm not logged in on my phone browser and it wouldn't open in my app
that’s weird. it’s actually a pretty useful feature, but it’s odd they’d add it to old reddit before new reddit, considering it’s basically deprecated. maybe it’s just an a/b rollout and i don’t have it yet
Sorry, I think I didn't explain my self correctly. That feature it's a very old one, it has been on old reddit since I remember. It has also worked on new reddit at some point, see the screenshot below from a comment I posted 6 months ago:
"View discussions in X other communities" feature in new reddit
:
In old reddit it's accessible from the "other discussions" tab
how the hell did i use reddit for almost a decade and not know about that feature
it wasn't your poor explanation, it was just me being an idiot i think - i just assumed it was new
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.
4:3 - Jumpscare, gremlin jumps in from off-camera.
16:9 AI upsized - Gremlin hangs out awkwardly to the left of the characters for half a minute, then jumps in.
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.
Haha, yes. I spent 15 minutes trying to remember the term for the pan/zoom-to-reveal comedy effect before giving up and settling on a botched jumpscare.
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.
There's a video from the YouTuber Noodle who does a good job explaining it
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.
load more comments
(1 replies)
Ahhh... the much-fabled "uncrop" operation!
Almost as good as Enhance
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.
Exactly, the filmmaker knew exactly what the aspect ratio was and framed shots specifically for it, why would anyone ever want this...?
Ooo, maybe we can get a nice blurred copy of the picture to fill the edges of the screen, just like TikTok!
I feel sick even jokingly suggesting that...
load more comments
(2 replies)
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).
load more comments
(2 replies)
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.
load more comments
(1 replies)
I remember watching a great video about why this isn't a good idea.
Clicked expecting Noodle. It was Noodle.
Nice video. But it can't be the right one. There's nothing about AI in it.
I was more referring to how changing aspect ratios is a bad idea, not that using AI to do it would be a bad idea.
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.
There is no original film for voyager it was filmed on tape.
TNG used film so that can be rescanned but the original analogue broadcast is the literally the best quality we have of Voyager.
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.
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.
edit: this is what i was referring to https://youtu.be/bD_HyxHMHPo?si=-ktJz3GuBNHhn9UY
I was wondering the same last week, but for Buffy the Vampire Slayer tv series, which received a horrible HD release some years back.
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.
Voyager and DS9 were shot on video, not film.
That's why there's HD versions of TOS and TNG but not DS9 and Voyager.
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.
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.
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.
load more comments
(3 replies)
view more: next ›