Google have a new AI for robots. China has been making great strides in robotics. Just needs the software to improve a bit, so probably within the next 5 years or so.
You know what I see happening? I see people just completely turning away from watching internet videos because we won't know if anything that we see is even real anymore, so it'll be a huge turn off, and people will just get bored, and want REAL first-hand REALITY again
IMO it’ll be about the value of the content - not how it was made. Simultaneously, we’ll have a Whole Foods market - but my guess is most will be down with Mountain Dew straight to the vein, which is what shorts/tt/reels feels like anyway
I see a problem with AI testing in that everyone's testing prompts makes too much sense...Yeah, even a godzilla-sized Corgi walking through the streets makes sense, the AI understands what it needs to do and can easily find inspiration to realize it. You need to challenge it a bit more, ask for stuff that doesn't make sense. All these clips you showed, it's all just normal stuff that can be found in any old movie. Even a skeleton playing guitar is not difficult for the AI to bring to life. Come on, show same damn imagination!! You have the most amazing tool at your hands, and you're letting your own imaginations limit your use of it!
I am sure google is adding some kind of watermark in these videos, or other way of identification. They don't want to pollute their training material with already AI generated content.
Sadly nothing AI is available in the EU on launch - not even Apple intelligence. Hopefully it’s just a matter of time to meet compliance with GDPR, DMA etc
The creative applications of text to video never made a whole lot of sense with models up until now. I'm absolutely stunned by how fast researchers have been able to get text to video working so well. Great work Google.
I wonder if there's accidentally the ability of location setting? Given Google's giant dataset around location, especially with Maps, Earth and Streetview, how narrowed down and defined can you set the location prompt. Like if you just said London, you'd likely get a London inspired generation, but if you were more specific, let's say Oxford Street or Harrods, how well would the model understand this to set the scene? If they've accidentally included this, or if they can implement it down the line, this would entirely change the game.
Well imaged locations definitely end up being consistent, but it’s not perfect. Eg look at the SF ferry building examples in my vid; the tower changes a fair bit. I suspect they will if they aren’t training on street view and aerial imagery already.
@@spatialintel Having mostly experience with Hailuo lately, I stopped watching at the halfway mark, so probably just cut off before you talked about Hailuo.
I am disappointed in it. Videos are so bad lately. Used to be 2-3 megabyte now it’s just 1-1,5 megabyte. Quality has dropped massively. I hope google releases veo2 and have an unlimited plan. Will definitely switch
It’s fairly clamped down prompting wise. Can’t put in names of celebrities/artists/IP also has a hard time with weapons and other stuff you’d find in a modern action movie.
There are many things this could be used for, especially as the technology matures, but when I see this, I’m immediately excited about the future of virtual reality. I understand that this type of technology is still in its infancy, but when I think about what it will be capable of, combined with conversational AI and real-time adaptation, I feel like a kid standing in the distance, watching a theme park being developed. I might be a bit overly optimistic, but I think we're starting to get close. If you disagree, please keep it to yourself. Don't be a wet blanket. Nobody likes you. 😅
Google is winning the generative AI race, but OpenAI has already achieved AGI, which means saying Google is ahead is like beating an F1 driver at football and saying you're the best athlete.
No upload of your own image allowed for image to vídeo , makes hard to have any use for professional use. From the video professional , point of view, they need more resources
It’s funny I was talking to my friend Allie about this who observed she can’t get Veo or any model to create a sneeze - and I wondered if that’s because people cut sneezes out of their videos so it’s not in the training data 😂
@@spatialintel 😂😂 that’s interesting! I actually didn’t think about sneezing! Makes me think about what other human abilities ai hasn’t quite got a handle on yet
@spatialintel using CUDA cores, and to think about it I only have a Titan X pascal I can only perform so many matrix till my GPU is exhausted, it puts your entire system in an optimized state of process to its maximum potential CPU and GPU communication in overdrive
@spatialintel I'm about to change the world of processing, I'm testing my CUDA application in Blender, this application puts your CPU and GPU communication in overdrive x10 the only thing I have to fix is when you render in Blender with cycles I'm getting an error I should find a way around it soon, compiling for all NVIDA GPUs will be releasing in the store soon, note this works for all applications, I have to find why DirectX is crashing also works flawlessly in Unreal Engine testing all sorts of apps
Though Google may surpass chat gpt, they'll still have the issue of have they absolutely fragment all of their services making it harder to use vs most of what chat gpt offers is on one location. The latter makes it easier for a normie to use, and thats where the majority wil be, and that's who will win. Hopefully they get this right,
yeah sora def has a nice look out of the box - it's the physics that are wonky. hopefully they roll out a bigger/slower model that is closer to the OG sora quality.
If they had launched in summer it would have been the leader for a while; then again the Chinese models like Kling and minimax kinda stole that thunder already
Google's on top, if you ask me. I played around with their Gemini Flash 2.0, and just the ability to toggle censorship? That's an instant win in my book!
Man we don't need AI video we need AI clean my house and AI wash my dishes lol
Ha, why not both! Check out this company called physical intelligence - they have a laundry bot that’ll fold your clothes too
@@spatialintel See that's something I can get behind lol
You going to CES btw?
Nah, have no reason to go to CES this year - but we have to link up IRL in 2025!
Google have a new AI for robots. China has been making great strides in robotics. Just needs the software to improve a bit, so probably within the next 5 years or so.
exactly
2000s: hey check to see if that is photoshopped
2030s: hey check to see if that is generated by AI
2060s: hey check to see if that is a naturally generated human.
You know what I see happening? I see people just completely turning away from watching internet videos because we won't know if anything that we see is even real anymore, so it'll be a huge turn off, and people will just get bored, and want REAL first-hand REALITY again
IMO it’ll be about the value of the content - not how it was made. Simultaneously, we’ll have a Whole Foods market - but my guess is most will be down with Mountain Dew straight to the vein, which is what shorts/tt/reels feels like anyway
is that why hollywood does not make movies anymore?
sounds like a blessing in disguise, tbh
It will be hard to tell what's real and what's BS. Soon, we will need government approve news channels that's free from BS
Spot on. 😂
I see a problem with AI testing in that everyone's testing prompts makes too much sense...Yeah, even a godzilla-sized Corgi walking through the streets makes sense, the AI understands what it needs to do and can easily find inspiration to realize it. You need to challenge it a bit more, ask for stuff that doesn't make sense. All these clips you showed, it's all just normal stuff that can be found in any old movie. Even a skeleton playing guitar is not difficult for the AI to bring to life. Come on, show same damn imagination!! You have the most amazing tool at your hands, and you're letting your own imaginations limit your use of it!
I don't think so. Sora literally can barely generate a video of someone cutting a tomato. I think veo still wins
Yeah I agree. But Nebol, if you have some wild prompts you want me to try I’d be happy to do it!
fr, people never tried asking ai for a new color
Yeah generic as fk accomodating overffiting
@@16comic But then how can we see it, though. Also I'm pretty sure anything the AI produces is going to fall inside the human light spectrum
I have been waiting so long for ai to be this good
I am sure google is adding some kind of watermark in these videos, or other way of identification. They don't want to pollute their training material with already AI generated content.
Yes they are - Google calls it SynthID
OpenAI should have released the full Sora model instead of the lackluster turbo model they released
Agreed, even if OpenAI only releases the big model to the $200 plan. Seems like they just wanted to copy Runway Turbo?
They did and it's far again compared to Google's
Whisk is only available in the US, no EU yet sadly
Sadly nothing AI is available in the EU on launch - not even Apple intelligence. Hopefully it’s just a matter of time to meet compliance with GDPR, DMA etc
@spatialintel yeah most stuff releases a few weeks after US release
The creative applications of text to video never made a whole lot of sense with models up until now. I'm absolutely stunned by how fast researchers have been able to get text to video working so well. Great work Google.
Me too! I still want more than text2video, but damn is it far more useful when you make prompt adherence a lot better
"What a time to be alive!"
The vibes every time I rolled a prompt 😂 now watch the feeling go away in a week cuz we adapt so fast
Fellow scholars!
I wonder if there's accidentally the ability of location setting? Given Google's giant dataset around location, especially with Maps, Earth and Streetview, how narrowed down and defined can you set the location prompt. Like if you just said London, you'd likely get a London inspired generation, but if you were more specific, let's say Oxford Street or Harrods, how well would the model understand this to set the scene? If they've accidentally included this, or if they can implement it down the line, this would entirely change the game.
Well imaged locations definitely end up being consistent, but it’s not perfect. Eg look at the SF ferry building examples in my vid; the tower changes a fair bit. I suspect they will if they aren’t training on street view and aerial imagery already.
what about consistency? multiple shots with the same person /creature animal
You’ll get a good sense at 22:16 - surprisingly good at locking down the character but def not as good as fine tuning on someone’s likeness
This is getting CRAZY. Think about this in 3-5 years.... it will be no different from real life. Even now its hard to spot
Feels a year away from video fully shattering the video Turing test
@@spatialintel Its absolutely terrifying what they can achieve. I can't even begin to imagine the complexity to build such a system
Bro if there is a way to get i to remember characters & Locations its over, as well as 20 - 60 Seconds
1:18 - Why TF does no one test or talk about Hailuo? It's WAY better
I do talk about them at 9:33
Yeah it’s free and good
@@spatialintel
Having mostly experience with Hailuo lately, I stopped watching at the halfway mark, so probably just cut off before you talked about Hailuo.
I am disappointed in it. Videos are so bad lately. Used to be 2-3 megabyte now it’s just 1-1,5 megabyte. Quality has dropped massively. I hope google releases veo2 and have an unlimited plan. Will definitely switch
Nice to connect brother, I spent 10 years as a global executive at Yahoo! Nice to discover your channel friend... namaste love this content, gracias!
Welcome brother
would be nice if Veo 2 could be run locally, or uncensored... Wait, is it uncensored? I dunno, i don't have access to it lol
It’s fairly clamped down prompting wise. Can’t put in names of celebrities/artists/IP also has a hard time with weapons and other stuff you’d find in a modern action movie.
@@spatialintel AS USUALL WE WONT SEE THIS MODEL FOR A LONG TIME ESP IN EUROPE
I was excited to try it out but couldn't sign in, so sadly I signed up for when the general public are let in, :(
Yeah, that's a shame. I hope they open it up soon, too! Meanwhile try Google whisk and imagefx
What’s up with all the latent space useage.
samanemami.medium.com/a-comprehensive-guide-to-latent-space-9ae7f72bdb2f
There are many things this could be used for, especially as the technology matures, but when I see this, I’m immediately excited about the future of virtual reality. I understand that this type of technology is still in its infancy, but when I think about what it will be capable of, combined with conversational AI and real-time adaptation, I feel like a kid standing in the distance, watching a theme park being developed.
I might be a bit overly optimistic, but I think we're starting to get close. If you disagree, please keep it to yourself. Don't be a wet blanket. Nobody likes you. 😅
Absolutely. It feels like speaking worlds into existence. And I can’t wait for the convergence of 3D + AI tech.
So when do we get to use it ourselves?
Apparently Veo is coming to TH-cam next year; that’s been announced - videofx tool release no date shared rn
Google is winning the generative AI race, but OpenAI has already achieved AGI, which means saying Google is ahead is like beating an F1 driver at football and saying you're the best athlete.
This and AR coming together is going to get crazy
Akin to reskinning reality
Open AI took a massive L with the Sora thing
If this is the baseline now then in a few years eveerything will be on par with cinema. Scary to think about.
No upload of your own image allowed for image to vídeo , makes hard to have any use for professional use.
From the video professional , point of view, they need more resources
I asked about this and it’s def on the roadmap - i suspect they wanna make sure 31:53 doesn’t happen 😂
As an european, I can testify we indeed eat our soups like Sora showed
Yeah, but.
Google only gives you a few seconds Invide AI gives you 4 minutes plus sound and voice.
I think if invideo had api access to Veo their tool could be wild for longer form creation - it’d do the Veo promoting for u basically
It seems they have train with videogames and 3d model with unreal engine
Yeah seems like there’s a ton of synthetic 3d training data in there
AI community and gatekeeping name a better combo
great vid soon we will be able to prototype movie scenes instead of explaining eFin storyboards.
💯
I watch this and assume all of them will have these same features and ability soon.
Sora destroys itself. It's crap
Shame, given how hype the results from the bigger Sora model seemed
@@spatialintel so larger better model not this cheap sora your seeing none of you will be able to afford anyway
Thank you this is dope
Bro how is this creating videos so perfectly when most still image generations still get 6 fingered hands?
most SOTA image models are pretty darn good with fingers now -- still not perfect, but the hit rate is MUCH better than it was last year
Now we know why it took so long for Sora to come out. They knew it was garbage.
Hunyuan isn't free. You have to buy credits to generate prompts.
The model weights are available for free on GitHub, but you obviously need a machine local or cloud hosted to run it.
Not to be all per say critical…but when ai creates videos allow Humans to BREATH like the motion of breathing , then it will be 👌👌👌
It’s funny I was talking to my friend Allie about this who observed she can’t get Veo or any model to create a sneeze - and I wondered if that’s because people cut sneezes out of their videos so it’s not in the training data 😂
@@spatialintel 😂😂 that’s interesting! I actually didn’t think about sneezing! Makes me think about what other human abilities ai hasn’t quite got a handle on yet
Man I know your voice , but I can’t remember from where !!
Prolly from The TED AI Show?
now i can make my own movies ....
Yea Open Ai gone have to drop that price google is killing them
Google having their own TPUs is a huge advantage too
@spatialintel using CUDA cores, and to think about it I only have a Titan X pascal I can only perform so many matrix till my GPU is exhausted, it puts your entire system in an optimized state of process to its maximum potential CPU and GPU communication in overdrive
@spatialintel I'm about to change the world of processing, I'm testing my CUDA application in Blender, this application puts your CPU and GPU communication in overdrive x10 the only thing I have to fix is when you render in Blender with cycles I'm getting an error I should find a way around it soon, compiling for all NVIDA GPUs will be releasing in the store soon, note this works for all applications, I have to find why DirectX is crashing also works flawlessly in Unreal Engine testing all sorts of apps
Seems better than Sora, albeit somewhat marginally.
Feel bad for all of the 3d renders out there
Though Google may surpass chat gpt, they'll still have the issue of have they absolutely fragment all of their services making it harder to use vs most of what chat gpt offers is on one location. The latter makes it easier for a normie to use, and thats where the majority wil be, and that's who will win. Hopefully they get this right,
Agree. Google needs to be better at this. And not kludge it into some dev tool either.
Excellent work by Google. As long as they keep their politics out of their tech they will be even much better.
u have more control over prompt detail but it still looks ai and smeary and saturated with over color. sora looks better, but less director control
yeah sora def has a nice look out of the box - it's the physics that are wonky. hopefully they roll out a bigger/slower model that is closer to the OG sora quality.
9,99$ a month for Hunyuan video
That’s the saas service you can get the model weights free on GitHub, of course you still need to have the compute or rent it which is a lot cheaper
goodbye stuntmen
Ka-2-Loo...
AND apple Disney Cyberpunk dancing no the Júpiter qnd blabla AND athromoephic cars2060 with 5 dimensión
Good video !!👽More
OpenAI deserves the fall, I was so excited when OpenAI introduced SORA they should release it sooner
If they had launched in summer it would have been the leader for a while; then again the Chinese models like Kling and minimax kinda stole that thunder already
im glad Sora didnt cost an extra fee because it is TRASH, I cant believe I waited so long for it smh,
China has petabytes of CCTV video to train with
Which makes them great at emulating CCTV footage?
Wow! Not even sure I want to animate anymore lol, this ai made 10 times what I could do in a couple seconds lol, bye animating!
Keep animation - now just set your sights on creating a short or movie instead of a shot
@ is a movie just a cluster of shots? Lol
Wow
how can it literally be your moms dog?
Because I tossed real images of her dog into Google whisk and it’s reverse engineered images to the point that she can’t tell it’s not him
Hi. Please send help. I loaded up labs. I've become a MusixFX DJ nearly instantly. Help, I am stuck at the computer. It is too awesome. 😆 🤖
Lol
sora looks better lol
All AI video look the same. No creativity
Be upfront about your shilling.. what's your relationship with this?
Bet ya won't
This ain’t a paid video. If it was, Google would have me jumping through a dozen disclosure hoops before I even posted it lol
Google's on top, if you ask me. I played around with their Gemini Flash 2.0, and just the ability to toggle censorship? That's an instant win in my book!
I am SO glad ai studio exists so we can toggle those settings. Now if only they had a nice mobile app!
@@spatialintel I agree! Super excited!
15H
186 L
28 C
6,84K S
1 153 V
DEC 20 2024
Just tasting something okay thanks lol
YT studio dashboard wilding rn 😂
@@spatialintelLmao yeah, basically. I like keeping statistics for videos I enjoy, that’s all lol