Deepmind seems to be really really competent. I don't think that $100+ billion valuation for OpenAI will last for long. At this point OpenAI has nothing that is ahead of its competition.
@@budekins542yes it still requires hard work of editing, connecting each clip together, and making consistent characters. I made a simple, 3.3 minute music video of a merman story using free, 5 sec clips, took me 7 days to finish editing.
It's crazy how a lot of people can't tell some of the AI videos are fake.I seen a AI video of an old asian monk, talking about spiritual stuff,and people were thanking him! Saying thanks for the inspiration.Were in trouble.
Great video Matt! Can’t wait for these video tools to become better and more accessible! Both models seem to have their strengths and weaknesses, but VEO really seems to be a step up from SORA in general!
Just imagine in the future: you will have favorite movies that nobody else has ever heard of. Custom designed for your specific viewing pleasures. Man, I'm never going to leave home.
So your favorite streaming channel just cancelled your favorite show when you were really wanting to see what will happen next season. Soon you'll be able to generate the next season yourself.
If you consider that this is still in the early stages of development, the videos are truly impressive. Especially when you take a closer look at some details-did you notice how well the water physics are executed in the first video with the dog in the pool? That’s remarkable! Depicting water with real physical properties is probably one of the toughest challenges for VFX artists to recreate. I’m not talking about the appearance but the animation-how it interacts with objects or other states. That’s really incredible! I assume that by the middle of next year at the latest, we won’t be able to distinguish AI-generated videos from reality anymore. And then, all video platforms will likely be flooded with such content. You’ll really need to be careful about what you believe. Ideally, all such videos should come with a mandatory label (AI Generated).
Check out my merman music video, i used free 5- secs hailuo clips on those, the water looks so natural and the movements... hailuo has done that already.
I used minimax in my merman music video, look at the water movements. Minimax is far better than veo. Those are just from free 5 secs clips stitched togethr.
@kayeaivideos never heard of that but will have a closer look but to your explonation this tool seems specificly constructed towards Special Effects i guess?
@@micromedia26 Minimax isn’t specifically for special effects but has better physics, which makes it great for realistic water movements. I haven’t tried Veo, but I’ve already made good water effects with Minimax, so I think it’s better for that.
@@micromedia26 Minimax isn’t specifically focused on special effects, but I’ve found it handles realistic water physics really well. I’ve been able to create smooth, natural water movements that I think are just as good, if not better, than some newer tools like Veo. While Veo is getting a lot of attention right now, Minimax is still great for tasks like this. Just wanted to share my experience with a tool that might not be the latest but still performs really well for specific effects!
I feel like the coherency and physical understanding of these models will just steadily improve, so what I'm most anticipating is *length.* I can't wait for the day when we're not limited to 10-seconds-or-less on generations, required to string them together in hopes it keeps temporal consistency across generations. I want to be able to prompt for, say, a 4 to 5 minute clip, minimum (ideally with multiple coherent shots, maybe even multiple scenes).
Take a look at today's movies/shows. Often the scenes last far less than a minute with the camera view changing often. if the producers continue that design, then there won't be as much need for very long videos. Coherency is key in that case.
Matt I’ve been watching your videos and really enjoy them. I think the AI tech is mind blowing. The quality of this video is scary good. As the google guy said this tech is the worst we’ll ever have. Given the short attention span people have it really concerns me the potential that is coming to fruition to be able to manipulate masses very easily. Maybe it’s always been there but this new tech is going to make it even easier. Not as pumped about the future as I was. Keep up the good work.
it's incredible where we're at!! 2025 will be a crazy year. You can kind of see how the models "imagine" when it comes to how they place the items prompted in the video.
I was hoping it was mode advanced than SORA.. I agree it beats SORA but still Minimax is much better in animation things like Hamster, Unicorn or anything non-human.. Here we see VEO2 is just making still frames..
Great surprise seeing actual tests that aren't cherry-picked and really showcased the AI video agenerators success to failure ratio. Despite the video title, the contents show something pretty good. Not flipping amazing compared to the competition. I prefer reality to hype, so really liked this video.
I have been critical of Google's handling of gen AI in the past, but even I'll say they've been hitting it out of the park lately. Looking at Veo2, it completely demolishes Sora in almost every way. Sora has been a giant letdown in so many ways. Very happy to see others swoop in and spank OpenAI because it's what they deserve at this point.
Impressive. But it feels like many people forget that the most important element is missing: acting. The ability to make characters speak and express themselves believably. I wonder how far away we are from achieving that?
Telling the difference in these examples is quite easy - *good quality* real video is high fidelity, meaning lots of detail. AI video can't generate that much detail yet. AI video seems like really well upscaled 720p at best (or even less). If you watch on a small phone screen, it's much harder to notice. But watching on a bigger screen...
I got all the others right, but I thought the rhino video was real because of the bouncing donger, I didn't think the AI would add that part. The fighter jet was only obviously AI because I understand what the instruments should look like, and it looks like I'm having a stroke trying to read any of the gauges.
Feels like we are hitting the limit of text to videos using this kind of architecture. It's getting better but mostly visually, the actual actions are messed up. They need to be able to ground them into real physics motion.
Seems like they have real trouble with multiple noteworthy things happening at once. You can have interactions in the foreground in an interesting setting, but you can't have interesting background actions, too.
Been waiting for ever on the waitlist for V1 even!!! :( I've had image FX since the beginning! This looks amazing but I wonder what Googles resistance to allowing us to use our own image references is. I Cannot wait for this and everything AI brings to 2025!!
It's cool but I don't think you'll ever be able to actually upload an image and then have it animated due to everybody being so afraid of copyright. Great demo though Matt!
One day, we’ll be able to rewatch our favorite series and ask an AI, "What would happen if Walter White didn’t die?" And the AI will bring that scenario to life.🥹
i think making a theater movie will take longer than conventional cgi, we will be flooded with funny and amazing shorts the coming time. when it becomes more powerful it will be great for people with talent for story writing and directing, no more actors and no more producers, nice.
Is Meta winning in all domains compared to OpenAi? Bought Gemini today. An other day of testing claud-o1 pro-Gemini-co-pilot today. An other fantastic video, Matt! Sending our engineers to you platform. How are the Prompt Awards going? 🥇 🥈 🥉
Imagine being OpenAI, making everyone get hyped for Sora, then deliver it poorly by virtue of letting the competition surpass you, at 200€/month. What a shame. :'D
A threat to Hollywood? No, and not because of technology. There is no star-power in AI, which most people like more that the images themselves. But it will be a genre and empower lots of Indie makers.
You can completely tell that these are trained on videos. I know, obvious but is even more obvious that it's just trying to duplicate videos similar it has seen by putting the components together. I think we've got a while to go before AI can actually generate a video rather than copy with a few bells.
Ultimately these video tools and also proprietary LLM's are only as good as the API that lets developers hook them up to a workflow... if its all via a predesigned front end that needs a human operator... well it matter not how good they are since they can't really be automated.
The way you say attire and enveloped… are you AI? 😂 never heard it said that way before. What state are you from? If you’re even human! 😂 Thanks for all the videos on this subject 👌🏽
Hollywood will be in danger when one of these models is used to fuly generate a feature film that passes the Cinematic Turing Test: was this movie filmed or generated with AI? Also, that film needs to include scenes with humans and dialogue between humans. I'd give that maybe 8 years or so. Till then, this stuff is both cool and useless apart from advertising and maybe an occasional replacement for stock footage for establishing shots.
Veo2 seems an entirely different beast to Sora. Vastly better with the physics from what we're seeing in these early demo's.
That’s what having access to the biggest video platform on the planet gets ya.
Deepmind seems to be really really competent. I don't think that $100+ billion valuation for OpenAI will last for long. At this point OpenAI has nothing that is ahead of its competition.
@@RyluRocky Good point
@@johnzach2057 2025 is going to be the most interesting yet for AI Video progress
@@johnzach2057l mean, openai and all the companies have something cooking that they cant release it yet because of laws or that are not ready yet
This video saved me $200 per month. Thank you!
do you even think google's would be $200? i bet you it'll be > $500
@@hqcart1 lol i'm using it free wdym? you didnt join the waitlist 3 months ago???
In few years everything about film production will change forever
You mean like one year lol
@@TheoBrownMusic7 could be, every time we think its gonna be this long, they bring things earlier than that. AI progress is accelerated accelerated
There will be riots and protests also
It Is here already.
@@gRosh08 at this point its nowhere near actual film production lvl.
Very amazing. Sora lost but is still the OG in the business. Cant imagine what we are able to do in 20 years
google owns youtube. i imagined their video models would outclass the competition
I've been using sora. Absolutely insane
yo .... if this is the new journalism, Matt gets a 5 stars. No cap
So long Hollywood. Hello Indy movies.
It's not as simple as that. These A.I video generators can only make videos that last a few seconds.
Wont always be like that
@@budekins542almost every film is shot in a series of many different camera shots that last a few seconds
@@budekins542yes it still requires hard work of editing, connecting each clip together, and making consistent characters. I made a simple, 3.3 minute music video of a merman story using free, 5 sec clips, took me 7 days to finish editing.
It's crazy how a lot of people can't tell some of the AI videos are fake.I seen a AI video of an old asian monk, talking about spiritual stuff,and people were thanking him! Saying thanks for the inspiration.Were in trouble.
Great video Matt! Can’t wait for these video tools to become better and more accessible! Both models seem to have their strengths and weaknesses, but VEO really seems to be a step up from SORA in general!
Just imagine in the future: you will have favorite movies that nobody else has ever heard of.
Custom designed for your specific viewing pleasures.
Man, I'm never going to leave home.
wait until VR is so real you'll design your own complete worlds to explore...or 1 billion others too.
So your favorite streaming channel just cancelled your favorite show when you were really wanting to see what will happen next season. Soon you'll be able to generate the next season yourself.
Do you see the downside of that way of thinking?
@@jeffkingston67 Oh yeah, it will probably be the end of humanity as we know it.
Buuuut...
I dream of creating series like The Sopranos or Breaking Bad... all with AI... or making my own documentaries. It's just so incredible.
WOW...great Info Matt...jealous LOL...been waiting for access to this on Google for a while... Hardly wait.!!! Cheers and great report!!!
Showrunner would be great using Veo 2.
If you consider that this is still in the early stages of development, the videos are truly impressive. Especially when you take a closer look at some details-did you notice how well the water physics are executed in the first video with the dog in the pool? That’s remarkable! Depicting water with real physical properties is probably one of the toughest challenges for VFX artists to recreate. I’m not talking about the appearance but the animation-how it interacts with objects or other states. That’s really incredible!
I assume that by the middle of next year at the latest, we won’t be able to distinguish AI-generated videos from reality anymore. And then, all video platforms will likely be flooded with such content. You’ll really need to be careful about what you believe. Ideally, all such videos should come with a mandatory label (AI Generated).
Check out my merman music video, i used free 5- secs hailuo clips on those, the water looks so natural and the movements... hailuo has done that already.
I used minimax in my merman music video, look at the water movements. Minimax is far better than veo. Those are just from free 5 secs clips stitched togethr.
@kayeaivideos never heard of that but will have a closer look but to your explonation this tool seems specificly constructed towards Special Effects i guess?
@@micromedia26 Minimax isn’t specifically for special effects but has better physics, which makes it great for realistic water movements. I haven’t tried Veo, but I’ve already made good water effects with Minimax, so I think it’s better for that.
@@micromedia26 Minimax isn’t specifically focused on special effects, but I’ve found it handles realistic water physics really well. I’ve been able to create smooth, natural water movements that I think are just as good, if not better, than some newer tools like Veo. While Veo is getting a lot of attention right now, Minimax is still great for tasks like this. Just wanted to share my experience with a tool that might not be the latest but still performs really well for specific effects!
Having UA-cam videos to train it with worked!!
Love your crowd sourced demo approach Matt!
The golden age of movies and Tv-Series is coming. Goodbye Hollywood
Please create yoga videos which other video generators struggle so much to create.
I joined Google's Video FX waiting list a year ago and then Veo 1 and now Veo 2 and have never received anything.
I feel like the coherency and physical understanding of these models will just steadily improve, so what I'm most anticipating is *length.* I can't wait for the day when we're not limited to 10-seconds-or-less on generations, required to string them together in hopes it keeps temporal consistency across generations. I want to be able to prompt for, say, a 4 to 5 minute clip, minimum (ideally with multiple coherent shots, maybe even multiple scenes).
Take a look at today's movies/shows. Often the scenes last far less than a minute with the camera view changing often. if the producers continue that design, then there won't be as much need for very long videos. Coherency is key in that case.
The new patron of the arts is your own creativity.
Krea AI is still the King when it comes to the Most Valuable System to Rent to use
Matt I’ve been watching your videos and really enjoy them. I think the AI tech is mind blowing. The quality of this video is scary good. As the google guy said this tech is the worst we’ll ever have. Given the short attention span people have it really concerns me the potential that is coming to fruition to be able to manipulate masses very easily. Maybe it’s always been there but this new tech is going to make it even easier. Not as pumped about the future as I was. Keep up the good work.
it's incredible where we're at!! 2025 will be a crazy year. You can kind of see how the models "imagine" when it comes to how they place the items prompted in the video.
Nice!
I was hoping it was mode advanced than SORA.. I agree it beats SORA but still Minimax is much better in animation things like Hamster, Unicorn or anything non-human.. Here we see VEO2 is just making still frames..
I like the tech for fixing and helping editors and film makers… not just generating the ENTIRE thing.
Great surprise seeing actual tests that aren't cherry-picked and really showcased the AI video agenerators success to failure ratio. Despite the video title, the contents show something pretty good. Not flipping amazing compared to the competition. I prefer reality to hype, so really liked this video.
Woah, just what a year 2024 has been
For sure syria is free too
"Dog jumping from water like a dolphin" would have been cool.
Veo 2 seems to be video generation's DALL-E 2 moment. The next couple of years are gonna be wild.
I have been critical of Google's handling of gen AI in the past, but even I'll say they've been hitting it out of the park lately. Looking at Veo2, it completely demolishes Sora in almost every way. Sora has been a giant letdown in so many ways. Very happy to see others swoop in and spank OpenAI because it's what they deserve at this point.
Also keep in mind this is V2 and the Sora we have access to is V1 with V2 on the way... Hard to compare the two
Impressive. But it feels like many people forget that the most important element is missing: acting. The ability to make characters speak and express themselves believably. I wonder how far away we are from achieving that?
matt flexing his new google buddies
Corpo wars are bloody EPIC!
Wait.... I couldn't tell on two of those 4 ai vids
Telling the difference in these examples is quite easy - *good quality* real video is high fidelity, meaning lots of detail. AI video can't generate that much detail yet. AI video seems like really well upscaled 720p at best (or even less). If you watch on a small phone screen, it's much harder to notice. But watching on a bigger screen...
Regular video has not gotten to real yet, but this AI model is getting closer to the HDi Video we have now. Still cool.
thank u for pronouncing Veo correctly, veo = "I see " in spanish.
I'm just crossing my fingers and hoping that it'll be reasonably priced.
You should try the same prompts with Kling 1.6 It's much better than 1.5 and 1.0 it consistently follows my prompts and understands anatomy better.
Thank you for not click-baiting something that actually stunned the industry
I got all the others right, but I thought the rhino video was real because of the bouncing donger, I didn't think the AI would add that part. The fighter jet was only obviously AI because I understand what the instruments should look like, and it looks like I'm having a stroke trying to read any of the gauges.
Crazy , this is So good
Feels like we are hitting the limit of text to videos using this kind of architecture. It's getting better but mostly visually, the actual actions are messed up. They need to be able to ground them into real physics motion.
Seems like they have real trouble with multiple noteworthy things happening at once. You can have interactions in the foreground in an interesting setting, but you can't have interesting background actions, too.
Meanwhile, Midjourney is trying to get hands right!
Thanks as always, Matt. 💯
Hi Matt, I applied a few days ago, how do I know if I have access ? Did they email you ?
Cant wait and at least we know its gonna be a "Sora Killer"
IMPRESSIVE
Good morning
Nice video
Upload them in a Google Drive and make it available for everyone
Been waiting for ever on the waitlist for V1 even!!! :( I've had image FX since the beginning! This looks amazing but I wonder what Googles resistance to allowing us to use our own image references is.
I Cannot wait for this and everything AI brings to 2025!!
crazy, i signed up and in waiting list. I dont know what im doing but hell yeah
Not in UK :(
Don't you have to join the waiting list?
@budekins542 yes but there is no UK on the list :(
How about adding the URL you are referring to to the video description??
It's cool but I don't think you'll ever be able to actually upload an image and then have it animated due to everybody being so afraid of copyright. Great demo though Matt!
Chinese will not care
One day, we’ll be able to rewatch our favorite series and ask an AI, "What would happen if Walter White didn’t die?" And the AI will bring that scenario to life.🥹
i think making a theater movie will take longer than conventional cgi, we will be flooded with funny and amazing shorts the coming time. when it becomes more powerful it will be great for people with talent for story writing and directing, no more actors and no more producers, nice.
Veo and imageFX solved most of the hand/fingers issue
looking at the progress it will be in next 5 years we will have a proper finale for GOT
Not available in France yet 😔
Is Meta winning in all domains compared to OpenAi? Bought Gemini today. An other day of testing claud-o1 pro-Gemini-co-pilot today. An other fantastic video, Matt! Sending our engineers to you platform. How are the Prompt Awards going? 🥇 🥈 🥉
Everyone has seen it. Nobody has used it. Still waiting for someone I respect (like you) to try it on camera.
What do you think of the latest imagen text to image generator?
Matt:join the waitlist, Me: Waitlist? Oh, wait I live in Germany it's not available here 🙄
Veo might be the future, but will it take jobs from creatives? Or will it open new doors?
Sounds like Sora stole your lunch money…
This should give Hollywood a big scare
Thank you 🙏🏽
Waaaaaay better than Sora. Just insane!!
They need to make an exception to not generating real people that allows you to generate Will Smith as long as he is eating spaghetti
Kling is better than both of them.
Hallucinations and inappropriate contexts still prevalent, making it a curiosity rather than a tool.
It will change everything provided its available to the world....
Still a very long way to go but the progress is impressive.
I have back tracked. I stand alone without AI. I can see AI down the pathway and I can catchup with it. But I know it will happen again and again.
Imagine being OpenAI, making everyone get hyped for Sora, then deliver it poorly by virtue of letting the competition surpass you, at 200€/month.
What a shame. :'D
it looks like VEO and Suno working better when they have to creat reality stuff
A threat to Hollywood? No, and not because of technology. There is no star-power in AI, which most people like more that the images themselves. But it will be a genre and empower lots of Indie makers.
Minimax Hailuo is just as good and already available
Can it generate videos of UFOs over New Jersey?
You can completely tell that these are trained on videos. I know, obvious but is even more obvious that it's just trying to duplicate videos similar it has seen by putting the components together. I think we've got a while to go before AI can actually generate a video rather than copy with a few bells.
Any Credits or is it free ❤😅😮
Ultimately these video tools and also proprietary LLM's are only as good as the API that lets developers hook them up to a workflow... if its all via a predesigned front end that needs a human operator... well it matter not how good they are since they can't really be automated.
“There’s a waiting list you can join”, you’re making the assumption that your audience lives in the US.
rip Sora
~Soon we can fix season 8 of game of thrones!
Sora is so sad 😂
soon you ll watch news on tv and you wont know if the news are even real
The way you say attire and enveloped… are you AI? 😂 never heard it said that way before.
What state are you from? If you’re even human! 😂
Thanks for all the videos on this subject 👌🏽
Hollywood will be in danger when one of these models is used to fuly generate a feature film that passes the Cinematic Turing Test: was this movie filmed or generated with AI? Also, that film needs to include scenes with humans and dialogue between humans. I'd give that maybe 8 years or so. Till then, this stuff is both cool and useless apart from advertising and maybe an occasional replacement for stock footage for establishing shots.
Sora clearly did better than Veo in several of those videos, in which Matt incorrectly called Veo the winner.
why are you calling it vayo when its veo
Kinda if Veo and Sora results came from a fella on Fiverr
TLDW: Sora is better at edge creativity case, and Veo2 is better at everything else!
Let's be honest ai video doesn't do ALOT for us so stop acting like it's important