Yeah, Google Cooked. Veo 3 is Mindblowing. Welcome to the Future

Google just dropped the absolute motherload of AI advancements on us. They just had their IO keynote. It was like three hours long. I want today to be more about trying the new AI out rather than just talking about it. But to get you caught up to speed on some of the best stuff that we saw from Google IO, we'll start off first here with Google Flow. This is a new website where you'll actually be able to generate videos with consistent characters. You can using the brand new VO3 model. This model has native audio generation, so this is mainly what we're going to be testing out and trying today. Here are a few clips just to demonstrate how much of a game changer this is. Bigfoot here, check my breakfast. All right, Bigfoot here it. I think he might be British. Google AI Studio has a bunch of updates. Lyria Real Time Music Generation. Don't forget about the straight up OpenAI Codex competitor named Jules Beta. I mean look at this. Connects directly to GitHub. It's a straight up competitor of course. Plenty of Updates regarding Gemini 2.5 Pro. It now has a Deep Think option that is going to be coming soon and you can see some pretty absurd benchmark scores. This thing is wiping the floor with O3 and 04 mini. Oh, and how about a brand new image generator? Imagen 4 is now available in Whisk for free. You already know I'm doing some pretty sweet lemon prompts with it. Google also created a diffusion based LLM which they're calling Gemini Diffusion. This thing generates outputs in an insanely quick amount of time. Like a split second, a blink of an eye, you get a full output. And so far, if Google's benchmarks are to be believed, it's actually pretty darn good compared to Gemini 2.0 flashlight while being way, way faster. This one is on wait list though, but some folks in my discord have already got access. I'll link this down below, but this should be like a pretty good roundup of everything they announced in the keynote. I think this is better than spending three hours watching the keynote itself. There's certainly a lot of stuff from the keynote I haven't mentioned. Again, it's like three hours long, but I want to play with the best of it today, so let's get started. First with Flow. This is all powered by VO3, so you're getting insane amounts of detail. I mean VO2 already is like the best AI video generator out there. So VO3, it's obviously insanely good. They do have some pretty intriguing short films on Here, but like I said, I kind of want to get hands on today. Here is the real kick in the groin. Right now Flow is anything but free. They do have a little bit of a promotion going on, but it's 20 bucks a month after that for the smallest plan on Pro and you are only getting V2 with that inside of this new Google Flow for a solid $250 a month. For Google AI Ultra, you also get the new VO3 access. Access to the ingredients, which essentially is the consistent characters and the privilege to top up your credits. Yep, that's not unlimited VO3 access. That is a credited VO3 access at about, I think 80 videos per month, which is pretty terrible still for that price. But yeah, you get the privilege to top up your credits with this insane plan. At least they are offering some sort of a discount initially on it. It's still mad expensive. Well, for those of you who don't have this kind of money just lying around or don't want to open your wallet for Google, you're lucky I'm here today. Ouch. This one was painful. Not gonna lie. That's it guys. We're down to my last buck. It's okay though, cuz now we have access to Flow Ultra. And by the way, if you do pay for this, don't forget to go to your settings here and go VO3 setting. You know what, we might as well just do a quick comparison between VO2 and VO3. A man strolls through a lemon tree orchard. He's wearing a ridiculously tall pink sombrero. It's at sunset and fireflies are dancing around him. Send that off. That's VO2. And now let's try that with VO3. Brand new model. Alright, now let's swap to ingredients to video. Oh my gosh. We paid for this plan and we don't have access to upload our own files. We can still generate an image with imogen4, but I can't put myself in here yet. Dude. Oh my God, this Flow thing is so buggy I can't even scroll all the way down. First up, VO2, we've got the man strolling through the lemon orchard. We definitely see the fireflies, although they're kind of cartoony and not really what I would like to see. In all honesty, he is definitely kind of just strolling around though. Very relaxed. Definitely a lemon orchard. And that pink sombrero is massive and absolutely ridiculous. Obviously VO2 is already a stunning video generator, so it nailed this pretty well right off the bat. But here's the kicker with VO3, folks. VO3 not only nails the prompt, but it also has sound right off the bat. Just native sound generation. So take a listen and a look at VO3's generation. It's pretty cinematic. He's still wearing a ridiculously large pink sombrero. The fireflies are a little bit more realistic, kind of dancing around still in that lemon forest as well. And yeah, it nails the prompt with sound effects. So bravo to VO3. But I'm not surprised that it absolutely slaps. I want to experiment more with the sound, so we're gonna do basic text to video. A human kicks a robot down the stairs. A clown runs down the street sprinting fast. His shoes make clown squeaks as he runs like I'm wondering if we can prompt some of our own audio into the system. Green goop flows out of the faucet onto the family cat, which scrambles to escape. A giant tortoise at the zoo transforms into a mech robot and lasers the zookeeper. This is a great start to the user interface. I love the concept, I love the ideas that they're throwing around here. But for the price, I expected it to be a little bit more functional and definitely a little bit better featured at launch. Quick heads up for you, folks. IBM and I are giving away 300,000 Granite AI tokens. Stick around for the next 55 seconds to learn how to claim yours. On May 28th at 1pm EST, I'll be hosting a live session showing how you can use IBM Granite AI to turn simple prompts into publish ready work. Zero code needed. Every single registrant pockets those 300,000 free Granite AI tokens. And that is enough for months of experimenting with Granite AI. Complete any of the three Spotlight activities and you'll earn an official IBM Skills Build Digital credential plus a slick digital sticker for your profile. The session will be interactive. We'll have a kickoff poll that allows you to decide which finals week headache Granite is going to be solving. First live visit IBM Biz Matvid Pro. And most importantly, click Enroll so IBM can email your personal webinar Zoho Link. Get in. Now we've got limited seats, limited credits, and an in room AMA after the demo. See you on May 28th. All right, now we've got the human kicking a robot down the stairs. And this should have sound as well. Oh man, the sound makes it so much more awesome. It looks like they're battling it out. Kind of crazy. We've got our clown. Okay, that definitely worked. He's Sprinting down the street as fast as he can. It definitely does the squeaks. Like you can prompt sounds for sure. That's huge. Amazing. That's perfect. All right. The faucet dripping goop onto the family cat. Oh, he doesn't like it. He's trying to escape. Okay, that was pretty great. Oh, and another cat spawned in. You know what for the prompt that I sent it. I'm impressed. All right, here's the tortoise transforming into a mech. Okay, just kind of cut to a mech now. Trying to laser the zookeeper again. You can see in both of these. Does its own black bars to make it kind of cinematic. Did we prompt for that? No. Here we'll grab the rocket ship. A thunderous spaceship flies into the sun. Sure. Instead of a clown, a monster made of ice. And instead his shoes make ice. Sound effects. Sure. As he runs. All right, how about this video? Still looks like we can't play this one. Question mark. Okay. It has so much potential. Flow has so much potential. But man, I am not paying 250 bucks a month to use this thing. Just give us access to the VO3 API. Honestly, I think that's all what we would prefer at least right now. Unless you build better interfaces and tools. Yeah, it's kind of crazy how many of these do have. Failed. Generations. This one we got nothing on the side. This one nothing on the side. This is a whole prompt that failed both videos. Meaning again, the clown one didn't get our second gen. The faucet didn't get our second gen. Are we losing credits? We're already down to 8,000. Alright, we've got our thunderous spaceship flying into the sun. And if you remember, this is from an image reference. Does it have sound? Yes, it does. Well, that's flying away from the sun, but close enough. How about this one? Very similar. I guess the sun is kind of in the background of it. There he is. I like the sound effects. Honestly, the sounds are pretty cool in this. Now we have Iceman running down the street. That's pretty cool. Okay. Again with the black bars too, though. It's like, guys, we're losing out on our video data here. I can add black bars and post very easily. I would much rather not have them generate immediately. Okay. I kind of feel like I need to do mech tortoise again. Single shot transforms autobot style into a mech turtle robot and lasers the zookeeper into dust. I think that is far funnier. First person view pushing a pink humanoid robot into a. Well, it Falls down lemons, rain from the sky and plummet onto people's heads. Wide shot, New York City. Some of these generations, I mean, when they get to 99%, you got to wonder if they're gonna fail. Yep, there you go. Failed Generation Monster. So we only got one Ice Monster gen. Another mech robot video, though. Yo, Mecha Tortoise. Not exactly what we wanted. What is going on? It's such a fun prompt, though. I gotta keep going. Just give me more. Infinite Mech Tortoise. All right. Pushing the pink robot down the well. First person. Perfect. The robot is so happy, but there he goes. God, the lemons raining from nyc. Oh, they're just kind of throwing them in the air. Sounds like New York, I guess. But the lemons are not raining from the sky, I'll tell you that much. It keeps cutting scenes. Oh, my gosh. That's crazy. Oh, my God. Absolutely burns the dude. The tortoises are not good. Again with the black bars, though. Like, we're not asking for that. I don't know why it's so intent on making these, because you're literally losing out on your generation. Like, you're. It's generating less useful pixels for you. Another really interesting quirk. No audio whatsoever inside of the scene builder. My question is, why? Like, again, if this is supposed to be for storytelling, where is our audio? The audio is awesome. I gotta say, the audio is awesome. This is sorely missed in literally every video generation model. Now that I've got to try it. All right, I. Monster is here again. Oh, that's so epic. It's pretty good, man. Like, this is. This is so sick. The native audio right off rip. And we got a second Ice monster. VO is actually behaving. Definitely got those sound effects. Yeah, that's pretty good. That's pretty good. I like the screams. They just spawn out of nowhere. Someone said slash spawn. Lemons 8,000. Very silly. All right, here's our suggestive robot shot. That's actually kind of a complete failure on the model side here. Some fun robot noises, but other than that, kind of scary and suggestive. I think this is the superior generation. Dude throws a watermelon through his window. He's super angry man. And job interview melts into blue sludge. Interviewer starts eating it with a spoon. I'm sorry. I love weird generations. This is the kind of stuff that's fun for me to test with the models. Oh, and of course, the defaults put us back into VO2, so I just blew more credits on VO2 instead oh, my gosh. I really gotta start to learn to actually check this before I click the submit button. But it's pretty terrible UI design, if I'm gonna be honest. Here's the guy throwing the watermelon out the window. Oh, this is VO2, though. He's just kind of angry. Doesn't even get it out there. This guy. Oh, there you go. You got the watermelon out. Oh, he's getting another one. There we go. Like, honestly, I miss the sound so much. Like, having the sound automatically generate, that's really a game changer generated with VO2 instead. Even VO2, though, it's just the generations are hilarious. This looks like the weirdest job interview ever. Finally, VO3. Angry watermelon. This is the most awkward toss. It looks like he's gonna smash it on the ground. No. How about this one? Okay. What is going on? It's like trying to get all the elements in there, but it clearly doesn't understand reality. Like, at the perfect level, you know, because things are so buggy and weird. But it's a lot of fun. Oh, God. Okay, this is the job interviewer one. Well, I'm very results oriented, and I pride myself on the ability to meet all of your targets. Well, I'm very results oriented, and I pride myself on the ability to meet all of your targets. I'm just turning into Blue sludge. I'm very results oriented, and I pride myself on the ability to meet all of your targets. That's insane. Okay. Very, very cool. I love the. The audio generation. It can do full sound like that. Like it full talking characters. Probably not consistent voices over time. Definitely not consistent characters with voices, considering they have audio. Straight up disabled. Oh, my God. All right, we got two blue Sludge ones. Okay, the job is yours. The job is yours. What are we making? What are we generating? Finally, Blue Sludge time. You taste great. You're hired, dude. All right, let's try some, like, vibe prompts. Worst job interview for a clown ever. Horribly wrong. Broken red nose. YouTube video. My pet turkey is British. Yes. I'm gonna blow through all of my credits. We're down to 5,000. Super Mario picking his nose in a public library. Housefly gets a drink at the local pub and complains about life. These more vague prompts, I think they're just gonna be a little bit more surprising. You don't really know what you're gonna get. Octopus. Selling bagels to mimes. I am very qualified. I don't think that's the case here. I am very. Yeah. You know, I don't even know what to say. I am very. I thought she was conducting the interview because dude clearly is the one trying to be the clown. This is kind of horrific. Very qualified. I don't think that's the case here. Literally rips his own nose off. That's just actually horrific. Dude is straight up Voldemort. My pet turkey is British. So I recently found out that my pet turkey, he's. But that's. He's Morty. He's British. So I recently found out that my pet turkey, he's. He's Morty. He's British. Very strange. So she's just kind of announcing it. I did prompt it. YouTube video. This is a little bit more of a professional one. I think he might be British. What was it with the laugh track? Like? I didn't even prompt for that. Guys, you have to understand, I think he might be British. Like, what is this? He might be British. All right, now we've got Super Mario picking his nose. There he is. Oh, he's just kind of pointing at it. I like the. The music in the background though. It's a little Mario esque. He's definitely in a public library. Very busy public library. It's like, I do have a nose. Yes. Oh, we got another Mario video. Again, just kind of touching his nose. I like this one though, because it's animated fully and it's just kind of like this quiet public library and he's just sort of moving around. No doubt. It's Super Mario though. It's pretty impressive, man. It's cool. It is cool. No denying that. Here's the house fly complaining. Wow, that's crazy. This life is such a buzz kill. It made a joke. I didn't even prompt that. This life is such a buzz kill. He's drinking about this one. No complaining in this. All right. The Octopus selling bagels to mimes. Oh, it's floating. I like this one. This is a good gen. I don't know why it's in like 480p though. High quality. Yeah, it's not streaming in high quality, that's for sure. Editing. Matt, stopping in here to tell you that the next portion of the video is going to be testing out the consistent assets and characters for a little bit context. This was actually recorded first, but I wanted to swap it around for the video because I thought the audio testing was so much fun. And easily the coolest part about this new model. Drop alright, text to image. Let's make a lemon character. That is a dastardly. Cowboy. Okay, these are pretty good. Yeah, Imagen 4 is definitely a great image generation model. It's pretty much on par with Ideogram's latest model, I would say. I like that it immediately also gave him a firearm. We'll use this image. Alright, here's a pretty good old western town. I think we'll definitely use that image. Let's go ahead and start prompting our little 3D animated character story. First thing I'm going to want is a slow pan over a desert at sunset. 3D animation by Pixar. Then I'm going to swap to ingredients. And now we want to establish our main dude here, our Lemon Cowboy. So we'll add him in there as an ingredient. View from behind Cowboy Lemon walking in the desert. Dramatic Pixar style again. Close up of an empty XXX bottle in an abandoned Cowboy era town. 3D Pixar Animation yet again. And of course using the town as our reference. Alright, now we've got our Cowboy Lemon walking around. This one does not seem to have any audio, but yeah, he's totally strutting around. This is pretty great. Overall I'm happy with this generation. Love the physics, love the lighting. Definitely seeing him from behind. So it's like you don't get the reveal of the character just yet. The XXX bottle is also here. Unfortunately though, you know, while it is definitely in the town like we expect it to, with the consistent assets, the bottle here has two weird. I mean this is not a normal bottle at all. Sometimes you just got to throw it in the image gen first and then get a reference, redo the prompt with the same town of the background and then toss it through this scrolling downwards glitch though they really need to fix. Alright, so since we did like this generation though, I'm going to go ahead and add it to a scene. So now that we're in the scene builder got a pretty typical AI video editing layout. We can kind of scroll through. We've got our main preview in the center. It's all obviously in the web. You can trim and like I mentioned earlier, actually extend these. But I'm going to leave the full eight seconds. I think for this one you can also still generate in here, which is good. It's very weird that this one didn't generate with any audio. I feel like that would add quite a bit. Maybe if you use the ingredients to video. It doesn't generate any audio here. Let's just run a quick test. Yeah, you see this one was generated without ingredients and we definitely have audio. This one was generated with ingredients. Yep. No audio at all. So ingredients doesn't have audio right now. That's intriguing. Well, we've definitely got our XXX bottle in full force in that old town. Love to see it. That can also be added to scene. So far the interface isn't too bad. It's a little bit buggy. Alright, slow pan through the desert here as our establishing shot. We've got our gunslinger lemon dude waltzing around. I really wish it had audio for these generated clips. And then we've also got our thing. We're definitely gonna need some other clips to kind of make this feel a little bit smoother. To improve this, we want an American desert in the morning. Wide shot of a lemon cowboy leaning against a dried up tree. He's enveloped in the tree shadow. And again, we want like dusty, dry vibes 3D animation by Pixar. Obviously using our dude for reference. You know what outputs per prompt. I guess we'll do two. Might as well. It always yells at me and gives me this error saying that it needs to switch us to a compatible model. Very aggravating. It constantly is defaulting this thing to VO2 when obviously I want VO3, especially if I'm using, you know, ingredients, which requires VO3 and is not something you can use with VO2. Oh my God. And it just deleted my prompt. Okay, the flow editor needs some work, guys. All right, let's try this again. Don't delete my prompt. See, there it goes. It still gives me the error even though. Yeah, okay. And then it just defaulted me to. What is the point of the model selector if it doesn't work? There we go. Okay, so I've made a new scene builder. I've added this scene to it. How do you actually save this? That is the real question. Right? Okay, here it is. I go back. Go back to Scene Builder. It's still here. Refresh. Okay. And then go back to Scene Builder and there's nothing. Okay, so don't refresh when you're using the scene builder because there is no way to save your scenes or your projects like that. Very strange. I don't know why they have it done like that. Again, just some weird choices that I would not be making. I'd like to see this saved automatically. Or at least give me the option to save it on the side. Because as I scroll around and click through here, I don't see anything like. Again, great detail. Good character reference in here. Looks like real 3D animation. He's just kind of yapping. I Think this one's pretty good, maybe usable. He does some great facial expressions, but not for an establishing shot of the character, which is what I'm looking for right now. Alright, we're at dice roll four with this establishing shot of our Lemon character. You know what, let's grab this dude. I'll try him in Runway. Gen 4 references. Same exact prompt in Runway's situation. It actually makes you gener the images first and then use them as image references in Gen4. Which might honestly be better than just generating the videos directly. Especially when it's difficult to get that exact shot you want like, I know it wants to do like a fun cartoon Pixar style deal with this character and it's always got him yapping against this tree. But I'm not exactly going for that. Maybe my prompting just needs some work here. But already, right off the bat, I'm liking these frames generations better. And I can convert into literally any of these into video. Okay, what's pretty funny is also Gen 4 Turbo is making him kind of yap. I do like the scene, of course, a little bit more from this image generation. Let's try the full gen four model, not just Turbo. All right, now full gen four. Yeah, that's definitely more of what I'm looking for, I would say. Just kind of vibing out. Got a slow panning shot. It's interesting to see the limitations between these models. VO3 is no doubt really, really great. And I'm sure there are a wide, wide variety of prompts that'll be very, very much fun. I wanted to do something with the scene builder here. I really wanted to like actually make something out, but I'm starting to realize that we're still pretty limited. Alright, let's try three references in total. We've got our lemon character, we've got the old town and of course the bottle. Alright, we've got our serious lemon dude. He's walking around. Oh, there's the big beautiful bottle into view. Oh, wow. This one actually just failed, so not great to see. But this one did indeed work. We've got our Lemon Cowboy. He's walking around the abandoned town and he sees the bottle just kind of come out of nowhere, which is interesting. I mean, it's pretty close to the prompt. I still again, love the physics as he walks and I like the dust and everything as the bottle kind of comes into view. I wanted his like surprised reaction as the bottle came into view. I think it was just a little bit too complicated for the AI to understand exactly what I was looking for. Also another strange thing that you'll see this model do a create its own black borders. Like it's hard to see right now. If I move out, you can see there's actually a black border at the bottom here, but not on the top, which is definitely not something that we would want. Might need a little bit more fine tuning to eek things out like that. Alright, before we wrap this video up for good. I know it's getting very long. I want to show off a few generations I did last night. I burnt through all of my credits, by the way. So that's what, 130 bucks gone down the drain. Seriously though, with the audio generation, I had so, so much fun. And I really do have some serious conclusions on this. But first, we've got an alien getting a slushie. Totally just hanging out. Little bit of weirdness with the slushie. It's like there's two cups in there and they're kind of morphing. You can see right next to my head over here. This one was a little bit better though. Like, what's up with that sound effect? It's like infinite interdimensional cable. Like seeing Rick and Morty. Right? All these strange things are now possible. You could just generate them and prompt them. 8 seconds is so limiting. Imagine if you could do 1 minute, 2 minute, 3 minute long generations. Then you're getting into like mini YouTube video or shorts territory and you could just generate it all with the audio and everything. Obviously could do music as well. He's supposed to have a case of beer on his back, but it's like stuck to his back for some reason. Just weird prompting issues. Some alien magic. The real world knowledge here. I mean, that YouTube training data is paying off. Like that's clearly a Chevy truck there in the background. These are supposed to be meme videos. Goblin casting a spell of confusion on a cat. It's starting to generate things that work right off the bat. If the generation was longer, like we could start just getting consistent stories, full videos based off of one prompt. That actually makes sense. Very intelligent AI video model. You are now under my control. How long would it take someone to actually animate this with the sound effects, all the clothing and the lighting. AI video is getting absurd. Planes filling up the Grand Canyon with lemons. Very silly, very weird, but it can make it. I like the plane sounds on this too. So mind blowing. Some wizard crystal ball action. Meow, Meow. It even says meow on the shirt. It's just so silly. You can hear the fire in the distance too. He's got a great voice. Ah, yes. This ancient relic contains the wisdom of the ages. See, going into the launch of Flow and VO3, I was really expecting the killer feature to be that Flow Studio consistent storytelling finally available. I thought that audio generation was going to be a part of that. Turns out it wasn't. Easily the coolest part about all of this was just seeing that raw VO3 output. High quality AI video that is intelligent enough to include audio that really makes sense and really is. High quality audio. Good voices plus good visuals, only eight seconds long. That's kind of our limitation right now, but it's very much possible to extend it. I would love to see a model that can just take a prompt and generate a five minute video, for example. I think that would be a game changer. Full visuals, full audio. I mean, who knows, Like a long enough text prompt, you could really tell a story, right? I really think that's where the future is headed. I like combining individual clips together and I understand love like it requires a massive amount of VRAM probably to do a traditional VO3 generation with the good video quality and audio quality for five minutes. But that, man, that is the future. What's this gonna be like in two years? How about five years? We've went so far in such a little amount of time. VO3 blowing my mind. And I want audio in all of my AI video generators in the future because it is just way too much fun. Makes it so much more worth it. Even when dealing with user interface problems, bugs, glitches and a very expensive plan hype for the future. Thanks so much for watching. Tell me your thoughts down in the comments below. I'll see you in the next one. Goodbye.