How To Use The Top AI Video Tools

2023 is coming to a close and we might have just witnessed what people call the chat GPT moment of AI video and that's because in November we experienced major advances when it comes to consumer accessible AI video tools. So this is the perfect time to cover it all but hold up with comparing this to chat GPT because while it might or might not have the cultural significance that chat GPT had for society video is a whole different beast compared to written text. So when we talk about AI generated videos, you have to realize that we're talking about a density of information that is unprecedented when it comes to AI tools. And that's why this is really significant because as you know a picture says more than a thousand words. Well video is 24 pictures every second plus audio. So this stuff really matters and if you think about it every social platform except maybe Twitter ended up as a video first platform. There's a reason for that and trust me there's a lot to talk about here. We now have a open- source video model stable video diffusion. Gen two, the leader of the fact, lets you now selectively edit parts of an image that it turns into video. And Pika Labs literally just kicked in the door with a trailer that is out of this world. It's actually a bit too good. We'll talk about that. And that's why today I'll be giving you a comprehensive overview of the AI video space, including some insights that you might have never heard before. And by the way, yes, there's more video related AI tools than we discussed in today's video, but these are the major ones. These are the foundational models. Outside of that, you have many apps doing interesting things like voice synthesis, very slight animation of faces and humans where it looks like they're alive, although they're just being moved a little. And then you have some AI video editing tools like Dcript. But today we're talking about the major models that are going to be driving this

whole space forward. All right, so one of those would be Runways Gen 2. So let's start our little journey through the AI video landscape right here inside of Runway ML's web app. And if you follow AI at all, you will have heard of this. They were the first ones making major waves with AI video that was not terrible. I'm not saying it was great. All of this has massive limitations. But so did AI images a year ago. Am I right? And as you'll see later in the video, we're really getting there. But it all started with Runway. Early this year, they released their first foundational model, Gen 1. And when I say foundational model, you can think of these as GPT4, GPT 3. 5, Claude 2, Llama. Those are foundational large language models. Gen one and then later on Gen 2 are foundational video models and they offer a lot of features. But let me just tell you, all of these are not created equal. For example, I still find that the text to video, I don't really know of a situation where I would go towards that. It's just not that great in practice. Whereas the image to video, where you give it a lot of information to work with, actually works surprisingly well. Matter of fact, if you've been following this channel closely, you might have caught on to all the B-roll that we put into our videos to freshen things up a little bit. And all of those have been created with Gen 2's image to video. All right, so before moving on to some of the newest models out here, I just want to highlight two more features in here that are absolutely excellent. And you can't really get these with the competitors as of now. One of them is the remove background feature, which works surprisingly well. You can paint over a subject and remove the background. And it's actually better than, for example, the roto brush tool inside of Adobe's very own After Effects. And the one standout feature that was added recently here is this motion brush right here where you can paint in certain areas of the image and it only animates those. Now, this is extremely powerful because you can finally control the chaos that AI video can be a lot of the times. And it works especially well on bodies of water or skies. It's absolutely terrible at animating humans, though, just like AI images were bad with humans a year ago. I expect this to evolve over time, but as of now, don't even try to animate humans if their features like hands or faces are visible. Like in this example, the astronaut might work because his suit is more of an object. And the last thing that I want to point out is that you're often going to be limited in terms of export duration. So your clips are not going to be a minute long. Gen 2 still leads the pack here with 4 to 18 seconds of duration. That's really good. But just like many large language models, this is a closed model. So all you can do is use this web interface. You can generate free clips for free and then you have to pay for it. But it's a closed proprietary model. Okay. And

before we look at the next AI video tool, let me tell you about today's sponsor, which is Wiretock. They make monetizing your photos, videos, and AI art really easy by uploading it to their platform, and they do all the distribution for you, which is usually a lot of work. And with the premium plan, you won't have to do that anymore. And once you have the premium plan, amongst many other features, you get access to their brand new styles feature. The most innovative feature probably is creating your own styles. So, let me create my own style by uploading 10 images of various cats with hats in a cyberpunk style. That is all right. And after a bit of waiting, the style appears on the my styles. By the way, here you have many preset ones, but here you can create your own. And then when I upload this image of a baby robot, I can remix it with my cyberpunk style and it makes it more into a cyberpunk robot. Like so. And look, after running it a few times, I actually really like this result. Now, just be aware that this works differently based on the subject matter. So, you might have to play a bit. But the cool part is they're going to introduce a marketplace for these styles. So, you'll have a way to monetize your very own style or a super simple user interface to reapply your brand style to whatever you're creating. And as you can see, you could also face swap, upscale, and then upload it to the various stock sites all in one place. So, thank you to Wiretock for sponsoring this video. Now, let's get back to talking about viral AI video tools. Now

this is where we start talking about stable diffusion video. Might have heard of stable diffusion, the open- source image generator by stability AI. And now at the end of November, they introduced a stable diffusion video model that is, and this is the highlight, completely open source, which means you can run this on your local computer completely offline for free. And with all the coming updates that they're already talking about, this is the worst version of this we'll ever have. So, let's have a brief look at this. Plus, I'll show you all the different ways how you can use this today. So, what's important here is that they came out with two foundational models, okay? One is for 14 fps video, and the second one is for 25 frames. If you didn't know, if you go below the threshold of around 24 frames per second, it stops looking like video to the human eye. Somewhere around 24, a series of images looks just like fluent videos. So, you get two models here, one of them being actual video. But there's a lot of limitations here, and these are best talked about while showing you the different sites where you can actually use this. One fairly obvious one would be a hugging face. This one is usually the slowest. You can just drag and drop images into here or upload them, use your webcam, whatever. And if you've been on Twitter throughout the last weeks, you might have seen all these video memes where they take popular memes just like this and they use stable diffusion video to animate it. And look, this is anything but perfect. But just like I pointed out with Gen 2, you will want to avoid humans in this. It's just not that good at animating them. It's a whole different story if the entire scene is in a cartoonish style that works really well. But look at that. This is something, right? And it's fully open source. So this site might be your slowest option. So I have two more here for you. One of them would be the decoherence app. And the good thing about this interface is they allow both image to video and text to video. With text to video, they probably just hooked up stable diffusion XL to stable diffusion video. So it first generates an image and then turns it into a video. Example coming in a second. Well, that's something. But before you click off this video, just wait until the end because the third model that I'll be showing here is truly next level and that's where the space is heading here. But here is actually my favorite one as I've been playing with all of these. This one works most reliably. Plus, it gives you a level of control that the others don't. So, here in replicate. com, you get a full form where you get to customize several attributes. And if you don't want to do it, the default settings work just fine. But if you want, you can. You can switch it to the 25 frames model. You can tell it to maintain the aspect ratio or crop it to 16x9, just like the format of the video you're watching right now. Now, played with some of these, but honestly, I got some really wonky results when playing with the motion bucket. And bumping up the noise just made it chaotic. And there you go. So, as you can see, when I bumped up the frame rate, the video got shorter. So this generated in under two minutes but yeah the limitation of the length here is one of the biggest caveats and about this problem with the length of the clip there's actually a fix. This tool has been built by one person and what it essentially does is it takes the last frame of the short video clip right so it will take this last frame right here and then it regenerates a new video and stitches the two together effectively creating a longer duration even though you're just limited to a few seconds here. As you can see this rocket launch is a great example of an inanimate object that it animates really well. The background is pretty still. It just has to figure out the movement of the rocket and how the smoke underneath behaves. So again, avoid complexity in the images you put in. You will get rewarded with better results. All right, so this web interface is free and absolutely fantastic. But if you want to go to extra step, I won't go into that now. I'll just point you into the right direction. So one thing you could look into is this Pinocchio one-click installer that allows you to install various AI models. And concretely, the one you would want here is Comfy UI, which gives you a node-based interface just like the one I introduced you to in the video where we built a no code chatbot. And that really allows you to dial in all the details. If you're interested in me covering how to set that up step by step so you can have that on your very own machine locally and offline, just leave a comment below. But if you want to learn more about that, I would point you towards Oliviosari Kas. He's a boatload of Comi tutorials. So feel free to check that out. Okay, so far so good. We talked about two major players in this game, but both of them have limitations, right? One limitation is the length. Gen 2 is doing way better on that. And another limitation is the resolution. Some of these are just not that sharp. But no worries, I got you because there's tools for both of these. If you're not aware already, Topas Video AI, it's costly, but it's currently the best software to upscale videos with AI. And this is what really allows for rich detail in some of these videos because by default, they're all just a little soft. And Topas really lifts it to the next level. This is not sponsored whatsoever. It's just the best tool available right now. So, as you can see, this just brings the footage to the next level. Now, most of the times it's not night and day, but it's the difference between h this is okay and wa that's actually good. And that's the thing with a lot of video tools and equipment. It even works that way in the real world. If you want to go from an average smartphone video to a actually polished clip for social media, it takes a lot of effort. You might need a DSLR or mirrorless camera, a proper lens, you start lighting, but then every time you try to make the resulting video just five or maybe 10% better, it's twice or three times the work. That's why TV commercials have crews of 20, 30, 50 people. and the result is not 30 or 50 times better than your typical Instagram commercial, right? And that's because upping the quality once you're at a certain level becomes really challenging. I think the same thing will apply in here. And that's why we've seen so many improvements over the last year because again going from a very primitive level of quality to clips like this is actually a big jump. But now we're at the level where it starts

getting tricky and this is where Pab shows up with their 1. 0 model like Unannounced. And boy oh boy did it blow people away. They're probably the reason people are calling this the chat GPT moment of AI video because this little trailer that I strongly recommend you watch is impressive. There's just no other way to put it. But what I would say is after you watch it for the first time and some of the hype wears off, turn the sound off and watch it like this. And this way you'll actually get a feeling for how good this actually is. Because when you consume the final product, they put together oh so well. The music and the sign design in this are so well done. going to take it to the that it makes it look way better than it actually is. And don't get me wrong, this is absolutely incredible. We've never seen quality like this out of AI video. If you look at it closely, it's not perfect, but it's just so much better than anything we've seen up until now. Look at these monkey getting sunglasses and them actually tracking nicely. Plus, the reflections work, too. Look on these last ones. It actually adjusted his head so they fit behind his ears. This is really impressive stuff. And it's getting something like this robot, right? Not bad. And I'm sure all of these clips are cherrypicked, right? But the fact is, no matter how often you run a similar clip through Runway Gen 2, well, at least from what I've seen, you're never going to get a robot walking just like this, especially from text to video. But again, just as we mentioned before, you can see that no photorealistic humans are featured in here just because they don't work yet, and I don't see them working over at least the coming months. These features here in the end are all about editing existing video. And at this point, I should also point out, yes, Adobe had some similar announcements with Premiere Pro, but we're still waiting for that. They just announced them, but during their keynotes, they showed off exactly this feature where you could change parts of the video with AI and it would actually track to the subject just like this top quite perfectly tracks. I mean, honestly, no matter how closely I value this black top, it just tracks perfectly. So yes, it's a perfectly produced trailer with incredible sound design and all the clips are surely cherrypicked. But the fact remains, you two will be able to create something like this with a freely available model because what Pika did here is they raised at a $200 million valuation and from what they've communicated so far they don't plan to make this paid for now that is just like OpenAI did with Chat GPT on release. Now one thing that should be noted is that currently Pika is available only for Discord and this will be through the website and Discord at the same time. Oh, one more really cool thing about them. When you read up on the founders and the interviews, they clearly state that the mission of Pa Labs is to create consumerfriendly tools. They're not making this app for Hollywood. They're making it for creators just like me. And that makes me super excited about the future because everything you can dream up, you'll be able to turn into a video. And then if you combine things like GPT4 or GPT5 for script writing, you use AI images to storyboard and then use AI video to bring the storyboards to life, we are quickly accelerating towards a future where even teenagers are going to become storytellers at a level that we've never experienced throughout human history. And I'm not making this stuff up. Literally, the youngest member of the team at the AI Advantage, Philip, a 15-year-old from Warsaw, sent me this clip today, showing off what he created in just 2 hours. not even using the new Pika 1. 0. So now we could start speculating what happens if people actually hone their skills and spend more than 2 hours on a project. Well, it's just a new era and we'll have to adapt to the fact that people will be telling compelling stories with visuals that go beyond what your iPhone can record. Like today, an individual wouldn't really be able to create animated anime series, right? Like you'll have people bringing entire fantasy realms to life with this stuff. Think about that. So many people will want to recreate Lord of the Rings. And that is just on a narrative side. What will happen to the videography industry? All the people that shoot videos for social media. Well, I would say if that's you, the only lifeboat I see here is actually learning these tools and becoming fluent at them ahead of some of your peers that due to inertia will just keep doing what they have been doing. Because let's just say one videographer is hired to shoot a short Instagram reel promotional video for a beverage and then another videographer that actually uses AI gets hired for the same job. Now, what are the outcomes going to be? By the time this guy sent out his first idea for the clip, the AI powered guy will already have delivered three different storyboards as proposals. And if he gets a green light, all he needs is some product photos. And then he'll be able to animate them. And very quickly, achieve quality like this, which currently is just unthinkable for a solo creator. But with a little bit of AI magic and a few more months, maybe a year of development, you won't be able to tell the difference between real video and AI video. Just like we reached the phase where with photos, you're not able to tell the difference anymore. Right? I don't know if you've seen, but this AI generated influencer that sells a Only Fans account has been going viral all over social media. And if you're not looking out for the fact that she's AI generated, the quality here is so good that most people wouldn't be able to tell. Expect the same to happen for video. just that I feel like this will have a way larger impact on society just because video is the most dense media format we have and all the biggest social media platforms rely upon video as the main medium. Again, there's a reason for that and this is about to change it all. So, I can't be more excited for the future of this space. I'll be covering it closely. I hope you guys enjoyed this overview. In future videos, I'll go more into depth. I wanted to create this first video to give you a broad overview. If you have specific requests, leave them in the comments below. And if you want to continue your AI video learning journey, check out this video right here.