So... the videos showing the difference between the AI-tainted youtube version and the supposedly untainted instagram version are hosted on... youtube?
I think AI-"upscaled" videos are as jarring to look at as a newly bought TV before frame smoothing has been disabled. Who seriously thinks this looks better, even if the original is a slightly grainy recording from the 90's?
I was recently sent a link to this recording of a David Bowie & Nine Inch Nails concert, and I got a serious uneasy feeling as if I was on a psychedelic and couldn't quite trust my perception, especially at the 2:00 mark: https://www.youtube.com/watch?v=7Yyx31HPgfs&list=RD7Yyx31HPg...
It turned out that the video was "AI-upscaled" from an original which is really blurry and sometimes has a low frame rate. These are artistic choices, and I think the original, despite being low resolution, captures the intended atmosphere much better: https://www.youtube.com/watch?v=1X6KF1IkkIc&list=RD1X6KF1Ikk...
We have pretty good cameras and lenses now. We don't need AI to "improve" the quality.
Two root comments (so far) are focusing on YouTube, but the article claims most of the AI was done by Will’s team, using AI to convert stills to video:
> The video features real performances and real audiences, but I believe they were manipulated on two levels:
1. Will Smith’s team generated several short AI image-to-video clips from professionally-shot audience photos
2. YouTube post-processed the resulting Shorts montage, making everything look so much worse
You can see the side-by-side [1] of the YouTube post-processing, and, while definitely altering the original, isn’t what’s causing most of the really bad AI artifacts.
Most of what YouTube appears to be doing is making it less blurry, sometimes successfully, and sometimes not. And, even with that, it is only done on Shorts.
I wonder if the fact that the original video was AI generated made the upscaling look worse than it would on a real video? Not that it can certainly be detected, but an actual video is likely different from an AI generated in ways that it seems like could lead astray their "computational photography" processing.
If I were a marketing person I would also make genuine images look AI generated for the free publicity. Nothing gets attention like mistakes or fakes. The fact that they aren't actually fake means there is no downside for WS and team. I once spoke to a social media manager for a large brand and he said they intentionally put typos in posts on a semi regular basis and it always results in more post engagement (people correcting the typo).
Seriously, who's idea was this? It can't be a money saving feature; surely it costs more to upscale all these videos than to just host the HD version.
And even if you argue it can be used only on low res videos to provide a "better experience", the resulting distortion of reality should be very concerning.
Theres not a single person i know in my life who will want this as a consumer. WHY does the world keep doing things that are so complicated and unnecessary.
Before AI was a thing, that same sentiment was true for almost every startup, especially the ones that had "get" in their domain name.
Yet here we are.
I find it hilarious that the Youtube spokespersons go out of their way to clarify that this is "not the bad GenAI shit that we know everyone hates but the good kind, you know, machine learning and stuff, you know, trust us"
I'm not sure about this specific instance, but AI generated movies will absolutely be the future, when you can create the exact shots you want with stability of the foreground, background, and characters, and edit it all together, it'll be an explosion of creativity just as with image generation currently.
To be clear, I don't think it'll be telling an AI to "create me a movie with X, Y, and Z" because AI reasoning is not there yet, but for the raw video generation, it's progressing steadily, as seen in r/aivideo.
The most incredible part about this story is that Will Smith is still a performing (and touring???) musician with any audience at all, AI or otherwise. I thought he was an actor now. Wut happened?
From a PR perspective, I wonder why YouTube is at the same time forcing unwanted AI features down people's throats[1], a move that many companies now do to drum up their perceived AI competence, but THEN at the same time, when asked, also downplaying this use of AI by splitting words.
The combination of the two confuses me. If this was about shareholders, they'd hype up the use of AI, not downplay it. And if this was about users, they'd simply disable this shit.
[1] I mean, they're sacrificing Google Search of all things to push their AI crap. Also, as a bilingual YouTube user, AI-translated titles and descriptions make the site almost unusable now. In addition to some moronic PM forcing this feature onto users, they somehow also seem to have implemented the worst translation model in the industry. The results are often utterly incomprehensible, and there's no way to turn this off.
My hardware/software or my eyes are borked because I cannot tell much of a difference between YouTube vs Instagram side-by-side. Gosh. If it is not my eyes, what are the recommendations? What is the top 1 (or 5) reasons I cannot see it if it is not my eyes? Do I need to upgrade my monitor? I have a relatively recent GPU but it is not a beast and I use a HDMI -> VGA converter.
The pictures, however, look god-awful! I presume the video is filled with stuff like these.
37 comments
[ 2.0 ms ] story [ 50.2 ms ] threadI was recently sent a link to this recording of a David Bowie & Nine Inch Nails concert, and I got a serious uneasy feeling as if I was on a psychedelic and couldn't quite trust my perception, especially at the 2:00 mark: https://www.youtube.com/watch?v=7Yyx31HPgfs&list=RD7Yyx31HPg...
It turned out that the video was "AI-upscaled" from an original which is really blurry and sometimes has a low frame rate. These are artistic choices, and I think the original, despite being low resolution, captures the intended atmosphere much better: https://www.youtube.com/watch?v=1X6KF1IkkIc&list=RD1X6KF1Ikk...
We have pretty good cameras and lenses now. We don't need AI to "improve" the quality.
> The video features real performances and real audiences, but I believe they were manipulated on two levels:
1. Will Smith’s team generated several short AI image-to-video clips from professionally-shot audience photos
2. YouTube post-processed the resulting Shorts montage, making everything look so much worse
You can see the side-by-side [1] of the YouTube post-processing, and, while definitely altering the original, isn’t what’s causing most of the really bad AI artifacts.
Most of what YouTube appears to be doing is making it less blurry, sometimes successfully, and sometimes not. And, even with that, it is only done on Shorts.
[1] https://youtu.be/Bx5GzIsmEBI
If there's code to stop AI from being trained on AI, I would like to have it from stopping me from seeing it.
Today on The Verge, GenAI upscaling in YT shorts. Yes, AI is here to stay, but I do hope the icky parts go away soon.
Seriously, who's idea was this? It can't be a money saving feature; surely it costs more to upscale all these videos than to just host the HD version.
And even if you argue it can be used only on low res videos to provide a "better experience", the resulting distortion of reality should be very concerning.
To be clear, I don't think it'll be telling an AI to "create me a movie with X, Y, and Z" because AI reasoning is not there yet, but for the raw video generation, it's progressing steadily, as seen in r/aivideo.
The combination of the two confuses me. If this was about shareholders, they'd hype up the use of AI, not downplay it. And if this was about users, they'd simply disable this shit.
[1] I mean, they're sacrificing Google Search of all things to push their AI crap. Also, as a bilingual YouTube user, AI-translated titles and descriptions make the site almost unusable now. In addition to some moronic PM forcing this feature onto users, they somehow also seem to have implemented the worst translation model in the industry. The results are often utterly incomprehensible, and there's no way to turn this off.
The pictures, however, look god-awful! I presume the video is filled with stuff like these.