New AI Use Cases You Have To See to Believe

It has been another week full of AI use cases that you could be putting to work today and this week, actually, the variety is amazing. The ones that I have today for you are really very different from the last few weeks. For example, an AI page with free stock images. Amazing. Or Mid Journey that finally released a tool where you can get consistent characters. That's the number one feature that has been missing across all AIR tools. And then we also have a few tips and tricks for things that we covered in the past, but new use cases come up, so we're going to cover those. So without further ado, let's get into all of this week's AI use cases that you could be putting to work today.

OK, let's start with Lumi, because I think this one is so convenient. And we talked about this before in the context of both photo and video, but AI is essentially the death of stock footage. And this is a fantastic example. This entire site specializes in giving you free images that look like stock, but they're all AI generated. But who would know? As you can see here on the homepage, none of these are really distinguishable. And all you need to do is download like so. And there you go. You can use these. Matter of fact, you can use these for non-commercial and commercial purposes. So essentially you can monetize them, do whatever you want. Now, it does link the creators on here, which I think is quite cool. And you could just add to this yourself for free. You just go up here, say contribute, and then you can link your portfolio. But yeah, if you need stock images regularly and you don't want to pay for them, there

you go. OK, next up, we have Inflection with their Pi model. And this one came out late last week, but our cutoff date is Wednesday because I always want to dedicate some time to testing each one of these tools. Plus, we spend a lot of time in editing, as you might notice. So every week we cover everything that came up to Wednesday. Then you get the video on Friday. So let's talk about Pi. This is a new foundational model. And as you know, we have a lot of these at this point. And again, it falls in this category that is just slightly worse than GPT-4. And yeah, we have many of those now. But this one is a little different because it's not trying to be your business assistant or your AI agent that gets things done. No, this is simply trying to be a friend. That's the impression I'm getting from it. And that's also what they claim. You can create your very own Pi. It's like a GPT inside of ChatGPT, but they're really not going for all of these business use cases that we talk a lot about on this channel. Because this is Pi, an emotionally intelligent AI. So you can try this out for free. Look, without even signing up, you just say next. It shows you a bunch of use cases that you could be prompting. And there we are. My first name is Igor. And now we go. Great meeting you, Igor. Hi, it's lovely to meet you. Nice choice. Now you can hear me. What should we talk about first, Igor? And look at that. Well, this is a new foundational model. As you can see, this interface is all about actually using it in your everyday life. And that's what this has been optimized for. And oftentimes a narrowed down model like this is better at those use cases than a general purpose one like GPT-4 that tries to do everything and does it really well. But even though this doesn't perform as well on the benchmark as GPT-4, it might be better at some of these. But that's just me speaking from experience, trying many of these products all the time. I hear you have an important conversation coming up. I can help you prepare for it by role playing. Can you please tell me what it is about and who you'll be speaking with? Oh, sounds like you're getting ready for an exciting conversation. So the focus of your talk will be AI use cases. Let's just say we're talking about Pi. Here are some key points you might want to touch on. So there you go. It reads out everything. As you can see, this is optimized for conversations. You have more recommendations here on the side. Just give this a shot. It's a really nice interface and it just speaks to you by default. And as you can see, what we predicted before is coming true. We're getting more and more specialized products where they're not trying to be everything. They're trying to do one thing really well. So if you want human-like interactions with AI, this is the one to try this week. And the next one is really, really big.

Because if you've been following the show or the AI image generation space, there has been one major flaw. And I'm not talking about text. Ideagram kind of solved this. We talked about this a few weeks ago. I expect this to get better in other tools over time too. I'm talking about character consistency. Because if you're trying to tell a story, if you're creating a comic book or a storyboard, usually humans really want to see stories with human protagonists. I mean, look at every movie ever. Usually a human is the character. And if that's not the case, then the dog that is the main character or the alien has human-like features. The problem with AI image tools is that every time you regenerated an image, even if the prompt was super precise, the human was slightly different. So you couldn't really do character consistency well. Now, there were a few techniques and there were some fancy solutions to this, like custom training a model. But Migerni now implemented something that makes this quite simple. It's not perfect, but it works really well. And the cool thing is you get all the Migerni styles. So here, Hades from our community came up with some images. He always hops on these things right away. They come out and within a few hours, he has some examples because he tested this. So this is where he took one image of me. Okay, this is from one reference image. And he created different portraits in this crazy, interesting style. And look, if you put these next to each other, this is the same character, more so than it has been before. Now, I'm not saying it's perfect. And I also look slightly different than in real life. Sure, but it is consistent across the images. I mean, this is not bad. Oh, and by the way, thank you so much for 200,000 subscribers. Really, I can't believe how fast we've grown. I mean, I did have my first YouTube channel 13 years ago. This is channel number five. And I'm so grateful that we get to explore all of these amazing tools together. To many more years and many more videos. I don't get to say this often, but your support means the world to me. And that I get to do this full time is really a dream come true for me. That's why I put in all the time and I try to make this as enjoyable as possible for you. In both the research and the recording and the editing. Thanks for 200k. And as a celebration, I want to show you something I've been preparing over the last few months. But first, let's finish exploring this tool because I'm about to show you how it actually works. So if you head on over to Discord and you have an active Mid Journey subscription, you can simply say slash imagine. Okay, so nothing changes here. Now you could prompt it and the key is saying CREF at the end of it. And what this wants a URL, okay? Basically, if I take an image of Charlie Chaplin here, I could just upload it and send it as a message inside of this chat. And then what I'm going to use is a specific prompt that I got from Hades here. But let's not forget the character reference feature here because now by including dash CREF and the URL of this image, which you can get by clicking it, saying copy link. And now all you need to do is paste this link. It will maintain a consistent character every time I reference this image. As you can see in the announcement post, there's a few more tricks you can apply like CW flag, where you can change the settings from 100 to a zero, where it takes away certain features like clothing or hair and just maintains the face. And the big disclaimer is that this works best for Mid Journey generated characters. Okay, so by me bringing in a picture like this, it's not ideal. It still works, but you'll get best results from real world people. But you know what? That's not bad at all. And the great thing is that I get to copy this last part right here. And every time I use that in an image, it's going to reference the same character and maintain his characteristics. So look at that. If I rerun this prompt three times and then run the same prompt without the CREF in the end, just with the link, you'll notice a significant difference. Every time I use the CREF flag here in the end, all these images are consistent. You can put them next to each other. And in my opinion, you're able to tell that, yeah, this is the same person across all of these. But as soon as you remove just that little flag, this is the result you get. None of these really look like Charlie Chaplin anymore. So a really powerful feature that we've been waiting for since a long time. Now, yes, obviously, Mid Journey is paid at this point. It costs $10 a month. But honestly, with this character reference feature, I think this might be so convenient to so many people because you could take your own images and remix them in any way, shape or form. Something that wasn't really possible with Mid Journey up until now. You would have got just different results that don't even look like you.

Okay, moving on to a very funky one. Okay, this is the most popular on Hugging Face this week. It's called audio editing. And it's very weird. It doesn't work most of the time. But when it does, it does crazy stuff. What do I mean by crazy stuff? Well, it turns the meowing of a cat into the barking of a dog. What? Bro, what are you talking about, man? I mean, who would ever need that? Please, challenge in the comments. If you can come up with a scenario where somebody would actually need to turn a meowing into a barking, let me know. I just can't think of a single scenario in life where that would be needed. But hey, who cares if it's needed? We can do it with AI. This is where we start. And end up. Okay, so played around with this a bunch. It's not that reliable at all. I mean, these presets work good, obviously, as they usually do. But you can do things like change musical styles or change one sound effect to another. And it maintains some elements of the original, like the cadence. And then if we turn this into an arcade game soundtrack. Again, this might not be the most useful thing in the world, but who knows? Maybe you create music and this might be great. You can have barking cats and such. I don't know. Just wanted to show it to you. Let's move on.

And now it's time for the thing that I wanted to show you off. That's the 200k celebration. I mean, it's a little celebration, but honestly, I couldn't be more excited about what I'm about to show you because I've been building this since, well, depends how you take it. Since November in the form that I'm going to show it to you and since over a year in a more extended sense. And it's really the AI Advantage Compatibility. The AI Advantage community, okay? And the images that I showed you come from there. And also this technique that I just showed you, this black line paid keyword is one of the almost 30 guides that we've been building here over the course of the last few months. And by the way, we're leaving out the sponsor for this weekend in favor of me showing you this. So let's take a few seconds. Let me give you a quick tour. It's not fully ready yet. There's just a wait list, but I'm ready to give you a first look here. So first of all, I assembled an amazing group of people, okay? We have our team. Then we had an alpha phase. Then we had a beta phase. And now we have our second beta phase, which I did with all the course members. So at this point, we have 250 people in the community. And we just did a lot of things which go beyond the course. Up until now, that was the flagship offering. But in here, you have the course, but also you have a Q& A area where every question gets addressed. We have a bunch of guides like this, which clearly show you what kind of results you'll be getting. And they range from chat GPT prompting to AI art, but also specific apps that do specific things. But also many other apps. If the community wants something, we'll likely build it and process. That's how this is set up. And here's my favorite part. We're actually doing weekly challenges in there, okay? So for example, just yesterday, we challenged everybody to create a custom song. But check out this one that Matthias created right away. That's it. These are his world famous lyric writing skills. I mean, this is super fun. But we challenge people to apply practical AI skills every single week. And the variety of answers we get there, and the way people go overboard and get creative, has just been super inspiring. I can't wait to show you more of this and more of the results in there. But this is the 10th weekly challenge that we're running in here. And we have an archive for all the submissions, but also all the winners. And beyond that, we run multiple events a week where we cover different topics, like how to run open source models easily. There's chapters, AI generated summaries. This is really where all my time has been going. If you ever wondered why I don't upload as much as I maybe could be, well, it's because most of my energy goes in here. So that's it. I just kind of wanted to give you a first preview. I'm looking to release this in early April. So if you're curious, there's a link in the description where you can sign up to the waitlist. But essentially, if you're subscribed to the newsletter or the YouTube channel, in the first half of April, you will hear more about this. So yeah, there you go. Exclusive little 200k preview. Thank you so much for supporting this channel. And now let's look at the next use case.

All right. So this app is called Avatar Dashboard. And it's not totally new. We covered these canvases that update images in real time and apps like Leonardo before. But this one is interesting for two reasons. First of all, it specializes in humans. So you can just move these around and you will be able to control the character very precisely, like so. And secondly, and take this one as a bit of a warning, this model that they use in here is very not safe for work, okay? So I have to be careful here on YouTube, but certain generations are just, yeah, very spicy. Okay, please, video editing team, please censor parts of this that are not appropriate. I never said thank you. And you'll never have to. So I don't know if you ever wanted an image of Borat partying with aliens in a very specific pose. Oh, God. Honestly, this might not be the most practical one, but it's a lot of fun to play with. I mean, I had a lot of fun playing with this. So I imagined you would have to. There you go. Avatar dashboard, you can try this out. All right, so next up, we have a really interesting app here.

And this opens a bit of a discussion that we should have here in the comments, and which is this app came out this week. Okay, it's called Devin. It's an autonomous agent that basically has a terminal, an IDE, and a browser, aka it can actually write code, execute it, go across the internet and learn across that. It's huge. It's revolutionary. If this really works the way they describe it, this is the type of stuff that everybody's hoping for with AI. This is one of the big productivity unlocks. I mean, heck, they claim that they put it in Upwork, and it's completing Upwork jobs for them. Now, the only problem here is this is not available today, right? And the whole premise of the show is AI use cases that you can put to work today. So I would just love to hear your opinion. This is obviously a massive use case, but it's not one we can use today. Where do we draw the line? I'm not exactly sure. Are you guys interested in hearing about things like Devin? Because this thing makes games, it trains AI models, and so much more. I'm not going to go super in depth on this, but this is really a future use case that made big waves this week. Maybe we could do a segment where we feature one future use case. Can't use it today, but it's good to know about. How about that? What do you guys make of that idea? Just let me know in the comments what you make of that idea. And if you haven't seen this, definitely go ahead and take the two minutes to watch this little demo video. It's amazing. This is the thing that people have been hoping for, and they seem to have built it. We'll cover this way more on release. For now, it's just demos, but it's impressive.

Okay, and I want to round this week out with a use case from last week, but a new application of it. And that is Cloudfree. We talked about how it's better in certain use cases than GPT-4. I even created a separate video on that, but there's a few new use cases that popped up, and I want to show you those too. So the first one is for code generation. If you're trying to create large blocks of code, Cloudfree is amazing at that. As you can see in this Reddit post here, Cloudfree created a simple neural network that trains itself on catching these parachutes. You can try that out under this link over here, where it just goes ahead and trains it. It places the target at different spots and tracks the data. As you can see, this takes a lot of code, and ChatGPT refused to generate all of this in one shot. So anytime you need a lot of code, Cloudfree is probably the one you want to use here. And here's another super interesting use case, and this is similar to an app that we covered last week, where it could locate different images from across the world. Now, this works with Google Maps data, but apparently it can take any screenshot from Google Maps and track it down and tell you where this is. Good luck. All right, so let's give this a shot. Let's go ahead and take a random image of, I don't know, Australia here, let's say. I mean, let's start with something that it should be able to get, like a more distinct coastline, like so. I mean, I don't think anybody watching this video would be able to tell what this is, right? Okay, it thinks it's Cyprus. That is not correct. Let's give it one more try. What about this part over here? Okay, it says Pilbara in Western Australia. Is that right? Yeah, that's the correct region. Not bad. Okay, let's do one more. Okay, so how about this random spot in Egypt? I mean, surely it won't be able to distinguish this. Southwestern Egypt. That's not entirely correct, but it did get Egypt correctly. So I can tell you from my testing here, it doesn't always work this well, especially if you pick something super indistinguishable, like maybe this square, it's just not going to be able to do it. I mean, it's not that precise, but it's not bad. I would say at the very least, it's worth a shot. Now, this is something that would have only been possible in movies a few years ago, right? Like a federal agency taking some satellite images and then running it through the software that tells them exactly where it is. Gosh, you have Google Maps and Cloud Free now that can do that, kind of. Pretty cool stuff. And that's really all I got for this week. If you care to learn about more recent use cases, here's last Friday's video. I do this every single Friday, where I catch you up on all the new ways you could be using AI. And that's really it. Subscribe for more and I'll see you next week.