Google's major updates

0:00 Welcome to your AI film news of the week. Now, before we get going, I want to give a special shout out to the Can International Film Festival for inviting Curious Refuge to speak at the AI Talent Summit this last week. The Curious Refuge and Promise team had an amazing time at the event. Thank you so much for the invite. All right, let's get to the news. So, of course, the first bit of news that we have to talk about is all of the announcements coming from Google IO.

0:46 More specifically, we're going to hone in on Google V3. So, we've been waiting for V3 for a while now, and it's pretty interesting because V2 was actually pretty dang good. And with this update, I got to say, I think V3 is the most intelligent AI video tool on the market. Now, V3 goes way beyond simply typing in a prompt and getting a video. It actually can do a lot of really intelligent things that other AI video generators simply can't do.

1:12 The first and most notable is it has the ability to generate voices and sound effects along with lip-syncing in the videos that you prompt. So you can type in a prompt, have them say whatever you want, and the lip-s syncing will be automatic. That's whether they're looking at the camera, looking to the side, it intelligently knows and will animate your characters to where everything matches. It also has the ability to upload a video and then you can type in a prompt to inpaint VFX inside of that video.

1:35 So, if you want to change any part of that video, you basically can use a prompt to make it happen. They also announced a referencing feature called ingredients that works very similar to cling elements or runway reference where you can take multiple objects, put them together and generate a video with everything combined. And then finally, they also announced first frame and last frame to where you could upload your own assets and basically interpolate between the two images.

2:21 There's a few other announcements for Google's new video generator that we'll talk about in just a minute, but you don't want to take my word for it. Let's take a look at a few community examples. So, we came across this video from REK that's basically a montage of V2 clips talking about the fact that they have the ability to speak now. Let's take a quick look. No more silence. Yes, we can talk. We can talk. We can talk.

2:21 We can talk with accents. Oh, I think that would be marvelous for Yes, it is very fun. Yes, it is very good. Very fun. [Music] Yes, we can talk. Yes, we can talk. We can talk. We can talk. We can talk. No, [Music] we can talk. So, as you can see, some of those shots are pretty dang realistic. Now, I do think the actual delivery of the voice could probably be improved a little bit, but good gravy, some of those shots, you would never be able to distinguish them from reality.

3:09 We also came across a great montage example from Angry Tom over on X with a few other examples here. So, first off, let's take a look at sound effects. So, Google VO3 was able to generate the sound effects for this wild off-roading adventure here. Let's take a quick [Music] listen. Okay, so I don't know what that's supposed to sound like, but I assume it's supposed to sound like that. So Google V3 is also incredibly good at physics.

3:32 So we have this film shot here of these two women looking over this cliff and get out of town. I mean the physics look incredible. The way the wind is flapping in the dresses, the way the handheld camera moves, the water coming off the cliff. Of course, you can really begin to break it down and see some areas that could need some improvement, especially in the fine details, but as a macro shot, and especially if you're looking at a smaller screen, I don't think you'd ever be able to tell the difference.

3:56 There's also this example of this animated character here. I'm one with nature now. Nature would prefer some personal space. Okay, so that's pretty wild, right? Like not only did it generate the voice and the character, you know, he's happy and then the angry bear has like an angry voice, but also the animation, right? Like you can feel the expressiveness from the main character with the joy and then you can also feel the anger from the claymation bear as he comes up and he has an angry face and he points.

4:41 There's a lot of really subtle storytelling elements that are happening with this tool. And of course, it would not be the internet without an ASMR video. So, we have this ASMR video of somebody clicking a keyboard made of candy. Okay, so I don't know what a keyboard made of candy is supposed to sound like, but it's probably a bit like that. It also understands lip-syncing for singing elements. So, we have this video here from Jared Louu.

5:14 Let's take a quick listen. We're going to light up the [Music] sky. I'm not sure I can go on. That is nuts, right? Like the drummer, if you look at him, it actually understands what symbols he should be hitting. And the guy playing guitar, it understands as he is plucking certain notes that he should be touching the strings. And of course, there's the dynamic movement up to his face, the expressiveness when he's singing. It really is incredible.

6:01 And the other thing that makes VO very unique from other AI video tools on the market is the fact that it's connected to a larger LLM. So clearly Google Gemini is powering some of the results that you get from V3. Case in point is this example from FFR where basically they just typed in a prompt of a man doing standup comedy and it added in this joke. So I went to the zoo the other day and all they had was one dog.

6:26 It was a Shih Tzu. Okay, that's actually like that's a funny joke, right? I think the timing could be improved, but it did a really good job with that joke. The big question is, is it an original joke? And the short answer is no. That joke was posted on Twitter years ago, and more specifically, it was posted on Reddit 2 years ago. But if you remember, Reddit and Google signed a deal last year where Google can train on Reddit's information.

6:49 So very clearly I think information from Reddit is getting put into the training data and it's able to intelligently generate videos that understand some of the posts from Reddit. And our final example here is one of my favorite. You've seen those terrible pharmaceutical ads on TV. Well, now you can create your very own in only a matter of seconds. And side effects may include diarrhea. Ask your doctor if it's right for you. If you make a parody pharmaceutical commercial, can you please send it to me?

7:36 I would love to watch it. All right, so we've taken a look at some of the community examples. Now, let's hop into the tool. So, in order to use Google V3, you need to go to the Flow website. We'll talk about Flow more specifically in just a little bit, but all you have to do is click the link below and then go ahead and sign in with your Google account. And from there, you'll land on this page.

7:59 That's the Flow TV page. And you'll notice that there's basically different channels where you can take a look at basically curated examples of what is possible with Google V3. There's all sorts of fun examples that you can poke around with. And of course, you can take a look at the prompt if you want and mess with it yourself and generate your own results. So, in order to get the most out of using Google V3, here's the big caveat.

8:22 It cost $250 a month. So, of course, that's going to be out of the price range of a lot of artists. However, they do have a deal for $125 a month for three months that you can take advantage of. So, just so you know, it's an expensive tool and for $250 a month, that gives you about 83 videos. So, you're looking at about $3 a video, which makes it one of the more expensive AI video tools on the market.

8:47 They do have a $20 a month plan that gives you way less credits. But if you want to just start messing around with it, you do have the ability to use that tool. So, in order to create a video, I'm just going to go ahead and go to the new project button. Go ahead and click that. And from here, we'll go to the menu. And there's a few different options. So, for now, we'll focus on text to video.

9:10 And let's say a middle-aged female astronomer stands alone on a hilltop observatory at sunset. yada yada. We basically have a long prompt. and go ahead and click generate. Now, Google's actually pretty quick. It takes about a minute to generate the results. So, you can see we have two video clips here, and they both look pretty darn good. I guess that's supposed to be like the Griffith Observatory in a different location, maybe something like that. Well, let's go ahead and select the video on the left.

9:32 And after you select it, you'll see we have this kind of new timeline view where we have the ability to rearrange our video clips. Essentially, it's kind of like a loose storyboarding tool. It's not meant to replace a video editing application like Resolve or Premiere Pro, but really a place where you can gather your shots together and see kind of how they fit together. And the other cool thing you can do is basically extend your video clips.

9:55 So, for example, with this video clip here, we can click the plus icon, click extend, and then we can type in a prompt, and basically we can say the woman looks at the horizon and sees a spaceship landing. We'll go ahead and just select one output here at the highest quality. and go ahead and click generate. And so after about a minute, it generated our video clip here. Let's take a quick look. So this is the original clip and now it transitions to the additional video clip here.

10:17 And I don't know, it just like randomly cut and now there's like an alien spaceship in the background. And I don't know how realistic it is. So clearly it's going to take a little bit of work to get the exact shot that you're looking for. And one thing I want to note is you do have the ability to download your video clip at 1080p. Now, it does have to be upresed because natively it's going to be in 720, but all you have to do is basically go to your video clip, click the download button, and then click on 1080p and it will upscale the video and then you can download it to your computer.

11:05 Now, we did come across one bug where we noticed that when you download the video clip in 1080p, it actually does not have the audio included in the clip. So, you might need to download the 720 version and basically compile it together if you want the sound effects to come through. So, naturally, we wanted to test Google V3 against other AI video tools on the market. And I think that Cling 2.0 is probably the best tool up to this point.

11:29 So, let's put them head-to-head. So, our first prompt is basically a woman wearing a kimono in a poker table and basically she says the words, "I told you not to underestimate me." Here's the result from Google V3. Remember, it generates the sound effects and lip sync automatically. I told you not to underestimate me. Okay, so as you can see, it did a really good job. I do think the overall aesthetics are a bit too contrasted and the lighting really kind of comes across a bit more like stock footage as opposed to a deeply filmic style scene, but not too bad.

12:13 And here's the same result from cling. Of course, all the cling examples are not going to have sound effects or audio because cling doesn't do that at this point. And here's the result that we get. You can see I actually think it's a bit more realistic when it comes to the physics, but who knows what she's saying. You would have to go in with the lip syncing tool and change everything. So, I think that in this case, Google V3 is definitely the winner.

12:35 The next prompt that we wanted to type in is a claymation animated man walking down the street. Here's the result from Google Via. Okay, so the physics look pretty good. The guy's face looks absolutely dead inside and the sound effects don't really match up with the person as he's walking, but it looks pretty good. And here's the same prompt inside of Cling. Okay, so yeah, that is a character walking down the street. I will say it looks a bit more like a Lego character as opposed to Claymation.

12:59 So, not the best result from Cle. I think Google V3 did a better job in this one as well. We have our third prompt here of a cinematic drone shot of the zombie apocalypse in an urban setting. And basically, buildings are on fire and people are piled up in the streets. It's World War Z basically. Uh here's the result that we got from V3. Okay. On a macro scale, it looks pretty good. I think the V effects there probably is some work to make it look a bit more composited in the shot.

13:51 And you can also tell with the zombies that there's just like some distortion happening where they kind of blend in kind of strange ways. So, I wouldn't say that this is silver screen ready, but it's pretty darn good. And the sound effects are really not that bad as well. Here's the same result from clean. Okay, so as you can see, this shot is just much different. We did prompt in for a drone shot. And it's kind of eye level, so I wouldn't really call that a drone shot, even though maybe you could have shot that on a drone.

14:15 There is some realism with the way the world looks, and some of the movement with the characters looks pretty good. In other cases, they're really just blending into each other. So, again, I think Google V3 did a better job. We also wanted to prompt in basically a teenage girl is playing an acoustic guitar and singing. And let's take a look at the result from VO3. [Music] Little bit of reverb there. That's uh not quite realistic, but it did a pretty decent job.

14:38 Here's the result from Cling. Okay, so again with Cling, you can see that there's just much more contrast and the lighting is a bit more flat because clearly Clling was trained on a lot of stock footage. So I think that the better result clearly comes from V3. And then let's take a look at our final example here where we basically have an action scene of a police officer running down a corridor yelling, "Wait just a minute." And I should note that all of these were our very first generation.

15:24 So, here's the result from VO3. Wait just a minute. That guy is way too winded. He needs to uh do some cardio work. And then also, I think he's running in stilettos because those sound effects are really sharp. And then here's the same result from Cling. So, you see the guy's running here and his boss in the background decides to start running with him. Okay. And as you can see, it's a pretty decent result here. Obviously, the guy is not saying the words.

15:48 So, I think VO3 still takes the cake here. But in terms of the physical dynamics and movement, it's about a tie for this specific shot. So, as you probably have been able to see, V3 is simply the best texttovideo tool on the market, and it's not really that close at this point. Now, VO3 also gives you the ability to take multiple assets and combine them together to create a video that basically pulls from different objects that you want inside of your scene.

16:13 It's called ingredients. Let me show you how to use it. So, in order to use ingredients, all you have to do is select ingredients to video. They will give you the ability very soon to upload your own image, but right now, unfortunately, you have to generate your own image. So, we'll click on generate image here. And we'll say a medium shot of a big black dog. And we have this adorable Labrador. Let's go ahead and use this one.

16:35 And let's create another ingredient here. And we'll go to generate image. And we'll say a wide shot of a porch in a rural Louisiana area. And we'll go ahead and generate that. Let's take a look at all these beautiful porches here. We'll select this one. Not too bad. And let's create one more ingredient here. And we'll say a wide shot of a young girl with short brown hair and overalls. And it'll generate a few different images here.

17:22 Let's go ahead and select this little girl. And go ahead and click use this image. So let's type in our final prompt here. We'll say a girl and her dog are sitting on the porch. We'll do two different outputs at the highest quality and go ahead and click generate. Okay. So now we have this video clip here. And yeah, it does look like that porch with the little girl and the dog. I do think that there's a bit more to be desired when it comes to the actual animation of the character, but all in all, it did a pretty good job and I would say it's fairly comparable to some of the other video tools that are on the market when it comes to using ingredients or references inside of your result.

17:47 Now, I wanted to test this against Runway's reference feature. And basically what we did is we took this image of a dog, this picture of a little girl, and this porch, and we asked Runway to put it all together, and it gave us this image here. And then we went in and animated it using Runway. And here's the video clip. You can see it just kind of slowly gets, I guess, closer to the dog, who I guess is the main subject of this video.

18:34 So, you can see the physical dynamics aren't quite as good inside of Runway. So, again, in this case, I think Google V3 did a better job. But there's a huge caveat to all of this, and that's simply the fact that Google V3 does not allow you to upload your own image assets yet. So from a film making perspective, it's going to be pretty challenging because you want to have consistent characters. So unless you create that character directly inside of Imagin 4, which is the image engine inside of Google V3, you really aren't going to have a consistent character or location.

18:56 So we're definitely on the edge of our seats waiting for Google to release that feature hopefully in the very near future. And speaking of Imagin, they came out with Imagin 4, which is their most advanced version of an AI image generator to date. And of course, we have to test it against other AI image tools that are on the market. So, in order to use Imagin 4, all you have to do is click the link below this video and make sure you're in text to image.

19:20 And we'll say a wide shot of a lone astronaut floating in space. Go ahead and click generate. And let's take a look at a few of our examples here. So, we have image number one here. It doesn't look very realistic, so not the best result there. We have this shot here of this astronaut, and presumably she's dead because she's not wearing a helmet. We have this shot here of a guy who is orbiting presumably Earth with an additional Earth in the background.

19:43 So, a parallel universe here. And then finally, we have this shot of this guy uh who's in basically I guess a screen saver of space. Uh so let's change this prompt and say a photo realalistic wide shot and we'll say of an astronaut floating in the earth's outer atmosphere and earth is in the background. So let's refine that prompt. And so we generated this result here and yeah, it looks pretty realistic. We have this one here.

20:29 I love how his shadow is like gigantic on the earth. So, and backwards. Not not the best result. The shot here, I don't know, not the best. And then we have this shot here, which actually I think is is pretty decent. I do think he probably should be holding on to something, but I guess hypothetically if you told me that this was a shot of this astronaut in space, I could potentially believe that the shot is real.

20:51 So, let's go ahead and compare three different shots here. So the first is a wide shot of an astronaut floating in space. From Image Info for we have the shot here. Again, it's not too bad. The second is a POV shot of an astronaut's hand reaching towards a shimmering circular portal and it created this shot here, which I don't know if it looks very realistic. I mean, we would have to see the way it animated, but you know, it's okay, I guess.

21:15 And then finally, we have a wide shot of an astronaut emerging from a portal into a lush alien landscape. You know, basically three shots kind of telling a quick little story. And we have this shot here. And he's clearly not emerging from that portal, but it's kind of what we're looking for. So, a decent result from Imagen 4. And we also went into MidJourney and generated the same results. Here's shot number one. I think it looks pretty darn realistic.

21:37 There's actually a lot of fidelity as well because you can natively upres directly inside of midjourney. It is a bit more stock photographyesque, but not a bad result for midjourney. We also have the hand reaching towards a portal, and this actually looks pretty good. There's some uh chromatic aberration around the fingers there. It's coming across a bit like a still frame in a V effect shot, so not too bad. And then we have the astronaut emerging from a portal on an alien planet.

22:01 And it's not imagining it as a magical portal. It's kind of thinking of it as a physical portal, like an actual opening of a cave. Uh, which is technically what we prompted for. So, you know, not a bad result from MidJourney. Now, I wanted to see all of the Google shots animated with V2. So, we have shot number one. Astronauts just hanging out in space and then his hand goes to a portal and then he ends up on an alien planet.

22:51 And you can clearly see there's a lot of work to do. You're going to have to work with the prompts. Hopefully, you'll be able to add in stylistic references and your own images very soon to help everything feel more cohesive. It was also really cool to see the announcement video that came from the flow announcement at Google where they actually invited Dave from our company and the instructor of our AI advertising course to create a film showcasing the new tool in action.

23:16 You can find more information in the link below this video. But more specifically, I think that flow is interesting because it allows you to just creatively get your ideas out there and on the screen in less time. Of course, you're always going to want to go in and refine and edit and do all of the sound design and storytelling that's necessary to make a film really compelling, but having the ability to generate your images and videos directly in an online timeline will hopefully help you save a bit of time and reduce the friction between coming up with an idea and actually seeing it on screen.

23:40 And of course, there was a ton of other announcements from Google related to search and a few other functionalities utilizing tools like Gemini. I'm not going to talk about those in this video because of course we're going to focus on film making, but if you want to learn more, there is a link below this video. And one of my favorite announcements that came from Google IO, and you may have seen it in the Hollywood Reporter, is the fact that they are investing in our parent company, Promise, to help us usher in a new era of storytelling.

24:04 Our company also announced investment from Michael Oitz, who is the founder of CAA, along with some other really big names in the industry. We're really excited to see where these partnerships take us along with the integration of Google's tools in our Muse AI film making pipeline. And just a heads up to you, enrollment for our June session opens up on May 28th at 11:00 a.m. Pacific time.