True AI Video is FINALLY Here & More AI Use Cases

So many AI announcements came out recently, but most of them are not usable, which is tough for this show because today it's Friday and yet again we're focusing on all the AI news that you can actually use. And this week has been AI Video Week. But hold up. Even if you're not interested in video production, this topic is significant for everybody. Because these video generators are heading towards a future where they're not just video generators, but world simulators. This is at least the vision of some companies like Runway with their gen 3 model that was announced this week that we'll talk about today. But there's also brand new things like uploading videos without audio and it just auto generates it for you. And we have a brand new anthropic LLM that apparently is dethroning GPT4O and it's also freely accessible. The fight is on. So let's look at all that and more in this Week's episode of AI news that you can use. Starting with Sonnet 3.5, Anthropic's GPT4O competitor. And this is quite a large topic, especially because they introduced this feature called Artifacts where it doesn't just write code anymore, it also runs it inside of an extra window in the interface for you. Things like this have been announced before, but no production ready ones have been shipped to consumers like you and me. This change Today, Claude Sonnet 3.5 is not just an announcement, it is available today for free. So rather than trying to speedrun through all the different interesting details here, and there's quite a few, I created a dedicated video for this that you can find on the channel. I'll link it in the cards up here. It should be available by the time this video goes out. And I'm just quickly show you a preview in two prompts like generate a portfolio website for a designer and make it more aesthetic. It doesn't just create the code, it creates this live preview in here with a little animation. Animations you can click all the buttons. This is interactive, this works. And you could build custom games in here. It can even generate sound effects for you in here. It is a wild capability and this model is available for free. Again, you can find a dedicated video just on this topic on the channel. But these are the first sparks of this agentic future that we're talking about so often where it's not just about assisting you but actually doing the tasks for you. Absolutely loving that development. And now let's move on to the next one which is Gen 3, the big announcement of this week. And look, this is news you can use. They usually only feature tools that you can put to work today. And as of the recording of this video on 20th of June 2024, Runway's new video generation model, Gen 3 is not available yet. But I really do feel like we do have to talk about it here because it relates to all the other tools we'll be covering this week. So all the announcements here happened three days ago, and they did promise this would be releasing to the public in the following days. And we took the opportunity here to show you some practical examples what these developments and what this leap forward actually looks like in practice. Because with all the examples they shared, they also shared the prompts. So to clearly show you how quickly the space is moving, we took all of these prompts, we ran them through their previous video generated model, Runway Gen 2, and then we compared Gen 2 and Gen 3. So here's one of the example videos you might have seen. This obviously impressive temporal consistency on the character. Plus the background in isn't funky either. It just looks real. And this is what the exact same prompt would look like with their previous model. Are you serious? Night and day. And this is why people get so excited about this stuff, because it's crossing a threshold where this stuff is becoming usable. A few more Gen 2 to Gen 3 comparisons here for you. This right here is the following prompt inside of Gen 3. And then we have the same thing inside of Gen 2, another Runway Gen 3 example compared to what you would get in Gen 2. Stunning and realistic versus probably not usable. One last one. And this really shows why people are saying this is almost at SORA level. It's comparable. It's a world generation machine where you can speak scenes into existence. Right? And then here again, the same prompt in the previous model, a big nothing burger. All right, all right, but I can already hear it. Igor. But this is not usable yet. Well, as they said days. It will be soon, I hope. But what is usable today is Luma Labs dream machine that we talked about last week, and last week it released. All of this is brand new when it comes to consumer technology. So people are taking their time to figure out what to actually use this for. And here are a few examples we found across the Internet. One of them would be these fashion ads. So you might have seen this with midjourney before, where people generate really extravagant and avant garde fashion designs. That would probably not be feasible here. They're doing the same, but with video. But to be fair, that's a bit of A playful approach, and probably not the real world use case that people are looking for. But the second one really shows the potential of these tools because here it has been used to create a little documentary. All the shots that you can see in here, the drone shots, they're not real recordings. They haven't been licensed, they have been generated. And if some of these shots would show up in the middle of any Netflix documentary, you or me, we probably couldn't tell if we're not paying close attention. And again, as mentioned, all of these are created with Lumalab's Dream Machine, so it's freely accessible. And you can use that today to recreate some of these shots. You have around 20 generations for free at the time of this recording. But there's more, and this is a fun one. People have just been animating memes. And it turns out that this is actually one of the best capabilities within Dream Machine. If you give it a picture, not the meme generation, but the fact that you can give it a picture and that it will turn it into a video, there's a lot of context within the picture, right? So you don't have to do intricate prompting. It will just bring it to life. And it works quite well. I mean, some of these are legitimately good enough that they could be real videos. And then obviously, there's also this idea of creating entire movies, and people have been attempting that too. Now, arguably, character consistency is not there yet. So from scene to scene, the character will look a little different, which makes this entire idea kind of unfeasible right now. Nevertheless, I want to share this with you, as you do have to account for the speed at which the space is developing. I mean, remember what we just talked about. We went from this scene to this scene within just one year. And it doesn't end there. The possibilities are actually quite deep. Some filmmakers are experimenting with this stuff and they're coming up with techniques like moving a camera around in the scene as you would with a steady cam. But then because you can simulate anything, they can just make that Steadicam fly as if it would transition to a drone shot. But all of this stuff would have been possible yesterday or a week ago. But the budget you needed to create a scene that is comparable to this with video VFX and traditional production would have been 10,000x higher than what this costs, maybe even more, because you have free credit. So essentially, it's free versus thousands of dollars, if not tens of thousands of dollars, to recreate something like this shot in real life. Sure, it would look more real. Sure. It would be probably more usable. And yes, I'm not even accounting for permits. And to round the segment on Lumalab's use cases out, I want to point out that they already announced some new features which are going to be essential, because right now it's only text to video or image to video. But what they announced here is editability. So you can alter the subject, you can alter the style selectively. You don't have to redo the entire shot, which is essential. And while this is great, over time, I expect something like Runway's Gen 2 capabilities, which are currently most advanced in the entire industry when it comes to consumer products, to ship to all of these tools, including dream machine and gen 3, once we get our hands on it. If you're not familiar, Runway has been adding tool after tool. And there are just so many ways to alter the video that you generate. They have video to video background removal, they have brushes where you can selectively change or animate only a certain part of the scene. And then they even added multiple brushes so you can do that to different parts. So with some of these improved generators, plus their tooling, you're going to be able to dial in a lot of details on these very capable models, which is something that has been unthinkable a few weeks ago because the base models were just not that good. But let's be real, the things that these models really excel at, from what we see here across Genfrey, Sora and Dream Machine, is these experimental and creative use cases. Hyper realism might be the holy grail because then you can do film like narratives. But for anybody willing to experiment with these world generation capabilities, I mean, this is a brand new idea. This is something we didn't have before. You couldn't just type in a few words and make them materialize in form of a video on your screen before. If you're willing to experiment with that stuff, this is an incredible time to be alive. Oh, and little human side note, if I look a little tired, it's because I am. My neighbors have been reconstructing for the past five weeks waking me up every single day. And I can record a stream during the day, so I do it at night, but then they wake me up again. It's a bit of a mess, but it won't stop me from delivering you all these new AI tools, tools that you could be using today, with a glimpse of the future sprinkled in here and there. Okay, on to the next one. All right, so the next thing I want to show you here is a suite of completely free AI tools and they also happen to be the sponsor of today's Video Corcel. So if you look at my screen here, this is a suite of generative AI tools that you can all use in your browser and they offer an API. So all of this could be built into other interfaces or applications that you might have. But let's look at these web apps here, because there's quite a bit here and as I mentioned, all of these are free to use. You can even check out more of them here here under apps. So first things first. Quite simple but very effective is this chat interface where you can switch between various models. So you could be using the advantages of GPT4O, like its expansive world knowledge to generate answers in here. Or you could switch to other models here, like cloud free Opus from its unique ability to write in a tone that is just different from GPT4O. You can also set the system prompt down here or use some of these open source models. But where this gets interesting is that you have multiple capabilities in one, right? You're not limited to Dall E3 when it comes to image generation. You just switch on over to the image chat and you already know what's coming here. We're going to do a classic cat with a hat and see what this comes up with by default. And the generation, speed, realism and editability of these is just far superior. What you get inside of ChatGPT by default. Again, all in one interface. But if you want to unlock most of that, you'll have to head on over to the Image Studio, which again is just one click away. So we'll just rerun our prompt here and in real time you can see this generating and you can already see the model itself using here. It gives you choices, right? So this is the standard one of Proteus, but there's multiple ones here. If you're looking for realism, you might want to use Dream Shaper for example. And as you can see, there's many other options here, like number of images, what aspect ratio you want it in, or we even have the ability to do generative expand or inpainting. No need to round trip to Photoshop or other tools. I can just drag my cat in here. I wanted to expand it like so, and there you go. It expanded the image, no problem. Same thing goes for inpainting. Super simple. I'll just roughly draw around the hat and then let's give the cat a giant Mexican sombrero. Just because I feel like that would look great on her. Yup. Turns out I was right. You can see that it maintains the style, so it fits the image here before, after. And this actually just scratches the surface. There's many more. Like an avatar tool where you can upload your own images and generate custom avatars from it. Or this character chat where you can talk to some of your favorite characters. And even a prompting trainer where you need to guess what the prompt here is and then you can compare the results. As I mentioned, all of this is freely available. Plus there's specialties that all of these tools can be easily included by an API to an existing application might already have. I just go to API, say add API key and then I can add functionality like inpainting to existing app. So if you want to try out any one of these tools, it's the first link in the description. Thanks again to Corcell for sponsoring this video and making this magnificent cat happen. Now let's move on to the next piece of AI news you can use. All right, next up we have Google Labs Gentype and this actually came out around 10 days ago. But hey, if we miss a tool and it is cool enough to be featured, you can be sure that if you follow this show and this show only, we will catch you up on the tools that we might have missed. And this is one of those. And look, it might not be relevant for most people, but I thought it was interesting enough to feature. This is one of Google Labs experiments. As I mentioned before, I'm a big fan of the Labs experiments. A lot of fun stuff in here and what you can do in here, you can generate custom fonts. So I'll type in news you can use in here and hit generate and it's going to create a custom font for me for news you can use. This thing is completely free. All I needed to do is log in with my Google account and look at that. It's creating a fully custom Alphabet for me. I could download all of these. Now the obvious feature that is sort of missing here is the background removal. So I could start using this inside of my video or graphics. You still have to do it manually, but we featured many background removal tools in past episodes of News you can use or you just do it the good old fashioned way in Photoshop where you cut all of them out. I barely finished my point and most of these images generated already. I have a fully custom Alphabet here that I can type in right here. How cool is this? Excellent. I'll just save this as a png. And look at what I just created in under a minute with AI. This is the type of stuff that makes me show up every single day. Research these tools. And while we'll keep doing this, this one in particular is not replacing creativity, it's enhancing it. Back in my commercial and corporate video production days, I would have absolutely loved this. I would have created custom fonts for most of my projects, depending on the client's needs. Or more of an everyday use case would be using this to communicate messages to loved ones. Look, you just type Happy birthday and we generate an Alphabet of something else than newspapers. Nevertheless, a fantastic little tool that you can try today. Okay, and next up we have Leonardo updates. Leonardo that we also talked about before is this all encompassing suite of various AI image generation tools. I think their mission is to be the Photoshop of AI image generation, but you can have all the tools in the world if the base model is no good. And this new model is not just good, it's actually great. It's called Phoenix. The main features are prompt adherence, meaning if you type something in, it's actually going to listen and not ignore parts of it, which is not the case with every image generator. The texted image is coherent. Let me tell you, from the testing that I'm about to show you, I would actually call it perfect and a few more changes under the hood. But what really matters here are the results. And I'm excited to announce that I actually have some real results here because we made a big change for news you can use this week. We're upgrading our workflow with multiple testers because I usually cover these tools and I play around with them, but I don't get to spend four hours playing with Leonardo. If I had to do that for every single tool, I couldn't bring out an episode once a week. But with the AI Advantage community being a thing, and with the AI Advantage team being 10 people strong, we can do things like send out Dominique, our midjourney lecturer and wizard within the community. By the way, here is his Instagram channel if you want to follow what he's up to with his midjourney creations. And DOM actually spent half a day testing this and reported back with these results. So first things first, let's talk about the new prompt enhancement feature because this new model really unleashes its abilities if the prompt is well written. Now, this can be a little tricky as it takes some skill and effort. So here are all the details that DOM ran in here. On one side, prompt enhancement. Off. On the other side, Prompt enhancement on. You can see the new prompt that it wrote. Instead of a cat in a fancy suit, it really fleshed it out. And here are the results. Much better. So it's not just a better model, but it's also more usable. Plus, it actually adheres to what is said in this prompt if you look at the details of it. But the big question is, how does this compare to Leonardo's previous flagship model, which was their Vision Excel model? Well, here on one side you have Vision xl. On the other side, Phoenix, same prompt. I think it is quite easy to argue that Phoenix is better in most ways. The proportions of the head are right, the details are better, and the texture in the fur aligns with the texture on the suits. Whereas with Vision xl, it looks like a bit of a collage of multiple things. I like it. Okay, next up, the text generation abilities. Ones with prompt enhancement off and ones with on. And as you can see, this text is sort of perfect, especially this one side with the prompt enhancement off. It's way better and needs no editing. Look at that. All four of these are actually perfect. When it comes to the text, it's a great tool to create stickers or other marketing materials. And then again, if we compare this to the older model, we just see more consistent results. And if you want to apply tools like this to business use cases, that's exactly what you're looking for. All right, so that's a lot of different creative tools, right? Image generators, video generators. We're going to talk about audio generators in a second here. But what do you even use this for in practice? Or how do you even practice? Well, I'm happy to show you one of the most interesting, unique, high effort, and fully AI generated things that we've ever done in the community. We hosted a super challenge where we challenged all of our members to create fully AI generated music videos, all the way from songs to the visuals. And I wanted to take a second to feature the winners here. The winner got three months of free membership. So, yeah, now we have a mechanism within the community how if you participate and if you're active, you. You get at least one chance per week to win a free membership. So here's the winner. Stavros and what he created is a music video. And listen to this. Based on a scholarly article investigating the speed of squid. Color changeability. Yes, you heard me right. A scholarly article about squids turned into a music video. Roll the video. Okay, I guess I'm learning something. I mean, let's be real. There's no Way you've seen something like this before. Love the uniqueness. Congrats again to him for winning. Here. Two more quick examples from the competition. Our very own NO Code automation teacher, Philip created this music video. Yes. And then one more which is a bit more melancholic and storytelling driven. So there you go. I just really wanted to share this with you. So much creativity with some of these and I just love to see it. When we get tools like this, you get the choice. You get to complain how AI is going to destroy creativity and there's nothing we can do about that anyway, or you get to embrace it and create things that have never been done before. To each their own. But as you know, I'm in the latter camp. All right, so for the last piece of AI news you can use this week, we have 11 Labs video sound effects, which also comes with an API. And it's exactly what it sounds like. You upload a video with no sound and it generates the sound effects. Now, something similar has been announced by Google's DeepMind, but it's not accessible, so we won't be extensively covering it here. Essentially, it's their video generator that also does audio that is coming soon, I guess. Pretty great considering we're just getting started. And the sound from video generation department. But this 11 labs tool you can actually use. They have a dedicated URL for it. Videotosoundeffects.com as per usual, all links mentioned here are in the description below. But let's just go ahead and upload something. How about I record with my phone and then I'll just do this without actually touching so there's no sound. Let's see how this goes. All right, uploading the video and let's see what we got here. Okay, so no sound effect, as you can see. What it obviously seems to be doing here is analyzing the image for anything that I can see. So you can see the keyboard, the computer, the screen, and it's generating keystroke sound effects. Plus some background foley of electronic static. Bit of a shame that we didn't get the sound effect. But if you need sound design for scene kickstarting your project with what it did here and these four different generations is actually very, very useful, although not 100% complete. Plus, as mentioned, they released an API for this. So you can build this into other applications, include this in workflows with a little bit of code. You could string together a pipeline where you generate images with something like Leonardo, then you turn them into videos with something like Dream machine or Gen 3. And with video to sound effects, you add the sound in and you could create automated music videos, explainers, documentaries, maybe sprinkle in a little bit of hey Gen, virtual presenter and use all of this as B roll, and you have a complete YouTube video, I suppose. Now, sure, none of these are fully there yet, but as you can clearly see, we're progressing week by week, and I'll be around to cover it all. I hope you enjoyed this video. Summer is starting, so I'm finally visiting my family and taking a few days off. But don't be afraid. There's going to be a new video every single Friday because companies keep releasing new apps, people keep coming up with new use cases, and I'm here every Friday to cover them for you. All right. I hope you have a good one. See you soon.