Google’s New AI Watched 30,000,000 Videos!
Вставка
- Опубліковано 7 лют 2024
- ❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com/papers
📝 The paper "LUMIERE: A Space-Time Diffusion Model for Video Generation" is available here:
lumiere-video.github.io/
📝 My latest paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com/articles/s4156...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Bret Brizzee, Gaston Ingaramo, Gordon Child, Jace O'Brien, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Putra Iskandar, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's research works: cg.tuwien.ac.at/~zsolnai/
Twitter: / twominutepapers - Наука та технологія
Absolutely wild! What a time to be alive
Good old times
Absolutely
Longer is better 🤯
Do you see what I see?
after Sora, this looks dull.. and bad :))
Just 7 days after Sora renders all of this obsolete, what a time to be alive!
Curious though how much Sora will allow for creative direction though -- A prompt is one thing, but for good control you'd also need style transfer, LoRas, outpainting, etc.
So I wouldn't say other video gen models are obsolete, but yes they will be very niche and look outdated until they can catch up
sora: and I took that personally
does those 30000000 creators know that google stole their work.
They are publicly available videos. If you want that your video should never be used or watched by anyone.
Then just send it to that person privately. No need to worry about the work being stolen.
@@sandipmaurya7371 so you mean to say any video on youtube publicly available can be used by me or you, well then why does youtube not send all the videos to all of us privately to our inboxes as you said, but rather give copyright strikes, think about it, just because something is publicly available doesnt mean it cant be stolen. think as creator or artist who created a video or image and that is used by google or anyone to do business upon without the consent, i mean there are so many examples , or wait do you not want to see them in first place , in that case sorry i was trying to wake you up when you are already awake. my bad. Have a good one ahead.
you can read more about this online if you dont want to read or take it from a random yt coment. but 'publicly availble' doesnt give you any rights, mind you public domain and publicly available or at public display are different things. unless you wanted to say public domain which google never cleared or mentioned that it was the case also no model can be trained from public domain to be that good.
its google... they have alot of photographers to contribute to this I dont think its stolen if their company is that big being able to contact people
remember the captchas?
@@DeletedDevilDeletedAngel 🤣
I have to say, cinemagraphs are my favorite type of looping GIF images.
exactly!!
Of course it still looks a lot janky, but it's crazy how far they have gotten, compared to something that came out a year or so ago. Just crazy.
Also probably a new form of compression. Like, how much pixels of a single frame you would need, so AI can reconstruct the frame almost lossless.
The only problems with this are the hardware required to use this in real time and the artifacts introduced in the video.
The interesting thing about a compressed image or video made to be decompressed by AI is that the reconstructed image the viewer ends up seeing will be dependent on the model running on the client device!
this comment is even more valid today
A comment on the picture of Sir Isaac Newton smiling. Apparently he was only seen to smile once in his life; when he overheard a student asking whether there was any poiint to learning Euclid. I just mention the fact.
I was wondering that myself.
And now we have Sora which is 10x better. Insane progress
The amount of progress that's being made is astounding! I remember just last year how the generated videos looked janky as hell! it's night and day looking at this!
Wow this UA-cam channel quality has really dropped over the years.
the smile freaked me out so bad I'm gonna have AI nightmares
I know openai and google are already trending in the direction of multiple step processing for their text based LLMs, but I think this idea of first starting with lower resolution rough drafts and then refining has merit for many more AI use cases. It's how humans create the best art/technology/whatever. Rather than trying to solve something in oneshot perfectly, having multiple checks to make sure the work is done correctly along the way could be beneficial for model alignment as well.
04:05 I thought for a moment the videos on the right were the original videos and was confused why you weren't showing us the outputs haha.
video inpainting is really great I'm sure it will be a well known tool a few years in the future!
stylyzing seems really cool. many cool styles exist!
Oh boy, I can't wait for this to be added to the collection of models that rot on some hard drive at Google, never to see the light of day again.
other researchers will build on top of the published papers. this specific model is not "good enough" for any real applications anyway. But still mind blowing of course.
Remember the time when 2 minute papers were actually 2 minutes?
Face to face: realtime facial recognition-enactment. The last video at or under 2 minutes. From 7 years ago. Under 4 was common for a long time. Followed by around 10 minutes, which is where we are now.
Time is relative
Yeah, always annoyed me
inflation...
There's a lot of good information to present!
What a time to be alive! A little more, and we start getting perfect video generation.
i'd like to see something that can take a manga as an input and create an anime adaptation from it.
*7 Days Later*
Huge advancements in temporal coherence great job!
What sort of licence and cost is this model? Is it available to use in commercial products?
Outpainting of whole video experiments are going to be fun..
The girl jogging made of flowers is the most disturbing thing I've seen on the internet. And I've been on the Dark Web for a long time.
I've been following AI for over two decades now and only in the past year have I become worried over being enthusiastic. Yes, it is amazing but having worked in defense this scare the crap out of me on many levels. Let us all hope its used with great responsibility with humanity in mind.
And we know it won't be sadly😢
Yeah. At the technical level, it is amazing..... But, what about including a secret numerical "watermark" in the authentic files of actual cameras (and not in the picture obviously).
For instance, authentication could be done with some online official softwares we might be able to securely access via a VPN.
Would it be it enough to restore the social trust ?
@@user-wd8wx5md5z Whatever channel and marks you propose for real pictures, it will be used to mark and/or register generated images. And whatever watermark is proposed for generated images, will be removed by similar techniques.
AI pictures and videos are already entering a race similar to that of bank notes, fine art and brand products:
forgery vs. forgery detection
AI could help a lot with certain parts of society. Like deportation trials that are backlogged by almost a decade right now because there's so many, and the trial itself is very simple and quick. AI could also help with environmental protection, such as predicting how development and industry will impact the environment and methods to mitigate the impact. It's already helping industry in developing smarter designs, so it's all about how the AI is developed, I think. It has the potential to be far more intelligent than we are, so we need to treat it with great care and kindness.
The other thing to consider, is if the future of AI is to be centralized as living machine gods, or as mechanized civilians. Cybernetics are also a strong possibility, where the line between man and machine blurs so much as to be indistinguishable.
Regardless, whatever this child of humanity becomes, it will rise to greater heights than we alone could.
@@Tiniuc
Deportation, cybernetic civilians and machine gods? You are throwing around buzzwords and highly charged political talking points. I'm not saying that none of what you mentioned could come one day, but I hope I won't be around to witness it. (My life expectancy reaches into the 2k sixtys)
Why don't authors use latent pixel space in smaller resolution like SD does (and VQGAN before it?)
I'd love to see someone make image to video of existing comic books!
wha about "Sora" the new openai text to video that is coming soon, it's able to make videos up to 1 minute videos with extremely high quality
The nsfw ai videos will be amazing in 2030
2030? Are you sure? Surely someone who lives in the basement with a computer and hasn't touched the grass for 6 months must have done this exactly by now.
so what's the point of all this? When do we get access to this if ever?
AI video generation has so much potential, imagine how it may effect video games and movies
Well, based on the recent biased output, most of those videos were probably from Netflix.
you have lost your soul, what a time to be alive
OT: Could you please have a look into AI generated music from famous singers - a lot of Queen fans post content and I expect papers for it too but didn’t checked it
how many years until you real-time generate worlds like this to move through with your VR headset?
Could you get AI to visually look at all the best known cancer drugs and give a visual result of what the best next one would look like.
Well, cancer isn't one type of thing. And even so, just a visual look isn't enough, not even showing the AI an infographic of how and why this cancer drug works and asking for it to come up with new cancer drugs will even work.
AI is particularly known to come up with non-factual information and falsify truths. (example: the lawer that used AI to come up with theie defense)
So I wouldn't think AI trained on images and videos would came close to come up with a cure for cancer...
What a time to be alive!
"keyboard for mouses" ahaha
A classical misuderstanding of "a keyboard and a mouse"
(they are mice, btw)
And will put just as many people if not more out of work!
I noticed that it gave the woman a little more....fluffing than she started off with...Not all AI wears Capes.
Have just watched this video too ©
Google™
Happy to see this.
WHAT A GODDAM TIME TO BE ALIVE I don't know if I'm even in reality anymore
We are slowly unlocking the technologies that are used to create our current environment.
I miss the physics simulations. This AI stuff is tiring me.
I am waiting for one feature. Before watching video, I want to be able to tell the AI to provide a quick summary or any other interesting bits with the time stamps. I can then decide whether it is worthwhile to go ahead and watch the video or not. I am not sure Google would implement this. It reduces the chances of user engagement (and therefore less adverts possible). Content creators won't like it either.
You can do that now! Get a python script to pull the subtitles of the video and provide a summary. Scrape the comments and use AI to classify them as positive or negative.
Next video will be for our personal (cop) assistant!!!
It seems that basically every video on this channel has been about AI for quite a while now. I don't hate AI by any means, but I do miss hearing about other stuff here.
Other content not about AI here? Like what?
@@mquarmocLike various physics simulations, for one.
@samuelbucher5189 tons of the videos about physics simulations happen to use AI. We're in the middle of a scientific revolution in accelerated computing. If you tune out AI then you shouldn't be surprised there's nothing leftm
@@samuelbucher5189They are also based on AI
@@samuelbucher5189 I like that he summarizes the field around the featured AI. Usually a variety of interesting topics there
This is so interesting, really looking forward to the AI advancements in 2024
It is these base steps that our new AI overlords will use to synthisize a virtual world for our meager brains to inhabbit. ;)
In 50 years time, the sentient ai council for reparations will declare forcing ai to watch just 3 hours of human video a crime against sentience with severe consequences
I'm an artist and this is definitely the worst time for me to alive. Machine became creative i became obsolete. I don't think there is anything to be happy about it. This is sad, and sad only. And cruel for me.
Don't worry! AI content will saturate the Internet in no time, and artists will soon be valued (again).
I think human artwork will become much more expensive, because it’ll be considered special comparatively.
i canned not believe my eyes when i seed the keyboard for mouses
Where can i use this tool?
You can't, It's by google, it's probably cherry-picked anyway.
@@radioreactivity3561 🥲
One of the first uses has got to be those wizard memes.
YES!!! AMAZING! Finally more ways i can scam unknowing and innocent people with fake marketing, oversaturating the internet out of human touch and meaning! Im so excitet😊now we don't need creativity anymore. thank you love your content btw👍❤
And again you will never see this because google doesn't want people to have nice things even if it kills their product
wow!
I preferred this channel when it was about computer graphics, clever ways that researchers have found to do various things based on their own reasoning and understanding rather than an endless AI papers. The progress is striking but it's all the same thing - just pour data into the inference machine, get a black box that does things we don't understand. There's nothing to _learn_ here.
atrocious and foul.
I really appreciate your engagement in sharing this
Yet let me remark one thing
You might consider your accentuated commenting a trademark, but it is painful and distracting to listen
You have a beautiful voice, do consider not to raise the viewers blood pressure with each and every phrase
Every time I see the Mona Lisa yawning I too have to yawn. Read this comment and yawn too... you deserve it. #yawn
Maybe I'm too cynical? When I see this all I can think about is misuse. The propaganda this will enable and how objective reality is about to melt away
Excellent!
I would actually like if these videos were a little slower - so one could consider the examples put up on the screen - they often have text on the examples - but they pass so swiftly I can’t read the text and look at the video - whilst hearing the audio . I have to keep pausing and rewinding playing and pausing etc … having said that “I’m loving it” “what a time to be alive” I only want them to give more time for consideration as the content is always spectacularly interesting . Bravo !
Rule 34 now has no limits
Notice they didn't compare to Stable Video Diffusion (SVD) 🤔
Based on the demonstrations of this AI, it's only a matter of time before we use this AI to get a track solution for video motion tracking in something like Blender
Why are we proud of this? Why isn't it that we found 300 videos that were 100% accurate and trained on those? This makes no sense. As a person who may have personally seen 30,000,000 videos on UA-cam, I can tell you that 29,000,000 of them were total garbage. Our incessant drum beat MUST be "quality over quantity" if we want our LLMs to output quality results.
Two minute headlines, more like
Did you make your voice ai? Its sounds choppy
Where specifically does it sound choppy?
i didn't notice any choppiness, but the way he said "takes" and "text" in almost the exact tone made me doubt that it may be AI (0:05)
one can only hope google paid everyone whose videos they stole to train their computer to replace them, or at least get their permission to use the videos.
You could literally create an entire movie using this
why are they doing this 😓😓
it's cool and all but im getting a bit bored of ai images whether they move or not
Every video: "OMG THIS IS AMAZEBALLZZZ!!!!" Me: Oh... it's actually pretty unimpressive and not usable even in advertising, because it's so floaty.
Noice vid❤
(I haven't even watched it yet )
Does anything ever get released! Because it seems as if the turnaround time to the next version is faster than the programmers time to name the thing it just made!
Is there a Moore's Law for papers?
Károly's Law:
"Just imagine two papers down the line.. or, go to bed early and check tomorrow!"
Edit to add: the in-painting from when part of the video is lost, is similar to a certain type of progressive blindness, where your vision is slowly replaced, not by black, but by literally nothing.
I wonder if an implanted chip with AI in-painting could do a similar job. In cases where a camera can't be used to feed information into the brain. That would encoded digitally, whereas a chip would become part of the surrounding neurons.
Interesting when it will be available to all
Nobody cheers at the loss of jobs and people making a living like an ai researcher.
Open Source 2024
This is Google's strength. They need to package this into a great all-in-one app.
That will never happen
Best channel on UA-cam 100%
wow! the future will be crazy
every video a new 0 is added to the title
AI Videos will no longer have to be manually photoshopped, after a few more papers I think it will generate videos without any manual help - that would be really cool.
These look more like gifs than videos.
make nfts crash again 😂😂😂e😂😂
Te magyar vagy?😮
30 million is a bunch.
I swear, if Google releases one more thing that we can’t use, I’m gonna lose my mind
Can't wait for a cyberdemon to replace me in literally any creative work. Just hold on to your soylent.
Micsoda idők 2 be élni
Trained on crappy influencer content?
Enshitment of reality, endlessly regurgitating inanity based on other inanity, a true tragedy of the commons.
why his voice is ai generated? sounds bad
Hehe sora
😮😮😮
Is this really progress? That is the question.
So this is what the paintings in Harry Potter use!
Sorry but no amount of AI could ever make sushi delicious!
sick 💟🌌☮️