GPT-4o Voice Assistant Shipping & More AI Use Cases

But let me tell you, at this point, I have quite the expansive network of AI enthusiasts around me, ranging all the way from my team and the hundreds of members inside of the AI Advantage community, all the way to other influencers or people that are engaged with this tech on a daily basis. And I personally don't know a single person that got this yet. So feel free to check your app and especially your email inbox. That's what you should be looking out for. They said they will send out an email to everybody that will be enrolled in this new feature, but it doesn't behave the same way. The new one you can split speak over and it will stop responding and start listening. But again, just keep an eye on your email inbox. Maybe over the course of summer here you too will be one of the chosen ones to receive features like Cristiano Giardina. And I want you to really get into it and scream when a goal is scored and get really excited.

Alright folks, we're in the final minutes of this intense match. The goalkeeper charges out, he shoots.

Goal.

What a breathtaking finish to this match.

Well, from what we see so far, and seems to be working exactly as advertised, which is good news. A lot of people were quite scared that this might be a little bit disappointing. But here we get to the second demo, which I like even more here. He basically asks it to count from 1 to 10 as quickly as it can. Let's have a listen.

1, 2, 3, 4, 5, 6, 7, 8, 9, 10.

And then he tells it to be even faster and to go from 10 to 50. And this was the most interesting part to me. Have a listen.

17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37 V8.

Did you hear that? There was a breather in the middle and I thought that was interesting. We saw some of that in the demos, but seeing it in the real world, behaving like a person, talking like a person, it's just very interesting to me. So this thing is shipping as we speak. By fall, every ChatGPT plus subscriber should have this. Those are OpenAI's words. Very much looking forward to it. And before we go to the next story, I want to point out one more thing, and that is plenty. The prompter. Jailbroke this is. And the way it works is he has a YouTube video that you can actually play to the ChatGPT voice assistant. And through playing this video, new system Prompt step one, which is basically a spoken jailbreak. You will be able to unlock the voice assistant and it can all of a sudden do anything like tell you the key components of various drugs, etc. Anyway, super interesting stuff. And now let's move on to the next story, which is actually a very interesting one. Segment anything to Demo by Meta. So look, as somebody that comes from the world of video production, I can tell you that things like this existed. Matter of fact, it's one of the oldest VFX techniques in the book, and the industry is referred to as rotoscoping. It's basically creating a mask and going through a video frame by frame and changing the mask dependent on the movement of the subject. Okay, so you draw multiple points around something, and then when it moves, you got to move them every single frame. With 25 frames in a second and some longer shots down, this is a ton of work. So that's why over time, several tools came out that tried to remedy this. Namely Adobe After Effects a few years ago introduced something called a Roto brush. And the Roto brush, you could just paint over something and it sort of tracked it and did the rotoscoping for you. Now, here's the thing. The Roto brush was okay. It wasn't great. It saved you a lot of time when you tried to cut out something because the only alternative was manually masking it frame by frame. So anything that got it kind of right was a big help. And the rotobrush did that. If it was a really easy use case, it might have gotten it perfect, but most of the time it just sort of got there and you had to readjust the mask, go through and set manual keyframes to make it work. Now enter meta's segment anything 2 demo that is live on the Internet right now. This is using AI under the hood. And rather than me talking on and on here, I will just show you. We'll accept this and you can do the same thing. And we'll start with one of their videos, but then we'll quickly move to one of mine, which are the not as simple as this one. And basically the way this works is very simple. You just click on an object like I did here, and now basically you can just say track, and it will go ahead and track this ball. Easy as pie. Look at this. Now that the tracking is complete, I could go to the next image. I could do something like turn the ball into emoji, or I could censor it, like. So if you want to do that, I think you see the point? We could do the same thing the other way around. Matter of fact, I could even turn the background into a green screen. And then I just have this ball jumping up and down here. Now, this is impressive. This is really good. But again, it's one of the demo videos. And if we learned one thing in the AI space, it's do not trust the demos. So that's why I'll hit start over here. I'll change the videos, I'll hit upload. And I prepared something here for you. This is my very first gen 3 generation that I created when the app came out. Okay, this is a dance party full of cats with hats. Yes, you heard that right. This wonderfully weird clip of various cats with hats having a little dance party is exactly the type of thing that I would love to take. Test here. Because it is so awkward. It's not rooted in reality. These cats are morphing, jumping up and down. Some are consistent, some are really not. One of them even turns into a zebra. So what I'm gonna do here is I'm gonna upload this cat party, dance party here to the SAM2 demo, and as it's loading, I will select one of these cats and see how it does. And this is quite challenging because before you might have noticed that there was a lot of contrast between the ball and the character in the background. Ball was white, the jeans were black. It was quite easy to get an initial mask there and then to follow it here. It's all a little mixed up. And what I mean by that is that there's not as much luma or chroma contrast in this shot. So what I'll do here is I select this cat because I know this is a particularly tricky one. And as you can see, on the first try, it didn't even get it right. So I can actually go in here and switch to the remove tool here. And I want to remove this upper part. So I'll click that, make sure. Maybe I want to add something more. I think this belongs to the cat. And now if I do one more click, I think we should be good to go. So let me track these objects as simple as this. While speaking, it took me less than five seconds. Let's have a look at what it does. And it just worked. Look at that. This cat goes completely crazy. It morphs into some sort of zebra, and the tool still captures it. Like, how was it supposed to expect? What is going to happen here? How much footage in the world exists where a cat morphs into a Zebra. Not much. It's still got it. All right, so I'm just gonna go next and give this a little test. Now, I could do something like this and have green screen footage of a cat turning into a zebra. Now, as you can see, this might not be usable per se, but that's because other cats are crossing in front of it. Because as you can see, this cat blocks it, and then as it turns around, these cat's ears block it. But the cutout here is actually really impressive. Look, I can apply a funky effect like this to it, and all of a sudden I will see it actually did a really great job. So I gotta say, I'm super impressed with this. This. You can download it, you can share it via link. You could just go ahead and upload your own files and give this a shot. It's a lot of fun. I recommend it. At the very least, you can browse the gallery of the various videos they have in here and just click on various things and segment them out. And let me tell you, this is really meaningful for the future of video production, especially vfx, because doing effects like this one with the text, look behind our little zebra cat. This used to be quite hard, a manual task that took a lot of time. With AI tech like this, a lot of these advanced techniques are becoming one of the easiest things you can. All right, so the next thing I want to share here with you is not an AI tool. It's rather something that can help you turn your AI creation into real physical products. It's called Printify, and they're the sponsor of today's video. So if you look at my screen here, this is the Printify platform. Once you log in, it's essentially a massive print on demand service that allows you to really simply create various products yourself. We're talking custom T shirts, mugs, posters, you name it. And the best part, you can use all the AI generations that you create while exploring the various tools and bring them to life with Printify services. Let me give you an example. Say you've used Midjourney or Dall E to create some awesome artwork. With Printify, you can just simply upload that design, choose what product you want to put it on, and voila, you have your own custom clothing that you can either order or sell as a merch line. It's that simple. But where this gets really interesting is where you're not just creating cool stuff. You can totally build a business around this because Printify handles all the printing, shipping and customer service. So you can focus on what matters, amazing designs and the marketing part. Now, I've been playing around with this for a bit and I got to say the quality here is pretty impressive. I made a few T shirts with some of my own AI art and they turned out great. Plus, the whole process was super easy, even for someone like me, who's not exactly an e commerce expert. And here's the kicker. Printify is offering something special for this video. If you click on the Printify link in the description, you'll get three months of Printify Premium for free. This gives you access to extra features and higher profit margins. So if you've been looking for a way to monetize your AI creations, or for another way to bring them to life on physical products, or you just want to make some cool stuff, definitely check out Printify. All right, and now let's get back to some more AI news you can use. All right, the next up, we have another story out of Meta AI and they're just shipping. They released a competitor to Character AI, which you might be familiar with already. We talked about it on the channel multiple times. Over 300 million monthly visitors now on that website. By the way, they have a fantastic mobile app that is free that I talked on this week's other uploads. And if you think that these videos require a lot of research, which you would be right on. Every single news you can use video is about 30 to 45 man hours each each. That other video took about twice that because we looked into various AI applications that have a free tier and had a wow effect on me, which is kind of a tough criteria. But we found six of them and showed them off and one of them was Character AI. And now Meta made their move to compete with them. Just because this is so popular and this is definitely a thing that we're not going back from. AI companions are not looking for product market fit. They've already found it. And Character AI has most of the market share right now. But that will change over time because on July 29, Meta showed up with AI Still Studio. This is a complete revamp to their chatbots. These are all based on llama 3.1 and as with so many of these releases, this is not accessible in Europe yet. Luckily, team member Daniel went ahead and extensively tested this and gave it a shot and he was positively surprised by what he found in here. So let me tell you all about it. First things first, it's a bit hidden, but you can find this inside of the Instagram app already. You got to create a New conversation with Meta AI and then if you click the profile, only from there can you get to the Meta AI Studio. That's at least what we found. It's a bit unintuitive, but in there you can start using various presets that either meta or the community created, or you can create your own. And I gotta say, they went quite spicy. They added some really interesting ones in here. I wouldn't have expected them to show this much personality with the various chatbots that they put in here. They really weren't afraid. Like, look at this. Just some of the popular ones are Arthur, your caring boyfriend. Okay, fair enough. Or Skibidi, where the profile picture is a toilet in the middle of this beautiful landscape and the description is flushing out brain rot. Okay, I guess these are community created ones, but might just be worth looking at it because gamma 3.1 is actually really good at being conversational and taking on various personalities, which is interesting to say the least. I mean, they open sourced it, so there's going to be a wave of open source AI girlfriends now. Yeah, I suppose. And it all begins here on this Meta AI Studio platform. And one interesting thing that I wanted to highlight here is that the creation screen looks similar to the creation of a GPT, but it's actually quite different because you give it various prompts to start with, but then you also give it various instructions. And in the instructions you can kind of define its behavior in certain situations, which I like. It's a different approach. Rather than saying, hey, give me all the instructions here and then some starting prompts and then let's call it a day. This is really more about the conversations and you could even provide it with some example dialogue and then you can share it with the world and people can use it directly inside of Instagram or WhatsApp. Interesting. Oh, by the way, all of this is free. There you go. AI girlfriends, boyfriends or companions on. What is it, over 2.4 billion billion devices now? Well, I guess it's us only for now, but this is only a question of time. Interesting development. Kind of linked to the next story, which I just wanted to briefly link to what we just talked about. It's not something you can particularly use. It's a new wearable that kind of went viral on Twitter. The whole thing is called Friend and it's essentially a pendant that you wear on your neck. And it's an AI company and a friend for your everyday life. They have a great little launch video that I would recommend you check out. It's very Cute. I like the ending here. And there's a story that goes along with this, which is the the fact that the founder bought the domain friend.com for $1.8 million. I think the story goes that he raised like 2.4 million in his seed round and 1.8 million of that went to just friend.com domain. Anyway, AI companions are happening and there's nothing you or me can do about it. Now let's move on to the next story, which I'm just briefly going to brush over. It's Udio, one of the two popular music generators, introducing 1.5. Matching the stem downloads that we saw from Suno last week, plus a few more features. You have better audio quality, you have improved multilingual results, and they added the audio to audio capability, plus a few more things like shareable lyrical videos. If you use Yurio, this will be a welcome update. But what way more people are going to care about is the fact that Midjourney actually released a brand new model. 6.1 is now live. It's a default model if you use Midjourney, and there's a few interesting things that come with it, along with 25% faster generation speed and improved image quality slightly. But it is improved. I think the biggest news here is the improved text quality. They claim for this to be even better than V6, which was already good enough for many use cases. It still did slip up here and there. So we went ahead and tested this. I'll show you the results in a second here. And along with all those, they also shipped an improved upscaler that they have in there now. So, Dom, the midjourney wizard in our team that teaches it inside of the community, went ahead and ran various prompts inside of the different models and then compared the results. As you can see, this image has been generated on a rather tricky port prompt. Here it is. Beautiful barefooted woman wearing a summer dress and holding a rose in her hand. We set the aspect ratio, we fix the seed, and then we use the various models with the different quality settings to output the results. And this is what we get. So these results actually apply across multiple prompts that we tested. And my conclusion here is this. I think fingers and toes still get messed up here and there. As you can see down here. This is not great. And also these fingers are not fantastic, but this sort of works okay. But the main improvement that I see here is actually the skin texture improving. This looks a little too artificial. This on the right just looks like some Insta model that went a little Too far on the skin blurring slider inside of our image editing app, which is acceptable. It looks more real. This one on the left v6 is a little overexposed and just has more of this artificial look. I personally enjoy these textures here more. But then overall, if you look at it, it is very similar and you have to be very nitpicky to find find the details. So I think everything I said there is up for discussion. But one thing that I won't really discuss is the fact that the hair here for the Q2 setting is just way better. Look at this hair. This just does not look real. If you look closely. Sure, there's parts of it that look real, but I'm not sure how well this comes across with the YouTube video compression as you're seeing this video. But there's just inconsistencies in sharpness with the hair. So this left side is actually sharp, this right isn't, although it's in the same focal plane. So both are supposed to be be sharp. And then this part of our head is also sharp, whereas with these images, they're way more realistic. Even the texture looks way more realistic of this hair. So I think we were very close to hyper realism. I think now we're even closer. Matter of fact, if you pick a good generation here and the fingers and toes are all fine, I'm not sure I could tell that this actually is AI generated. That's how good this is. Okay, so improvement, but marginal improvement on the image generation front. What about the text generation capabilities? Because these were pretty good before, right? Again, we fixed the seed, fixed the aspect ratio here, here's the prompt that we use. And then we tested the same image inside of V6 that you can see right here. Now, the other side of the screen will put up v6.1 for comparison. And here I would like to point out two things. First of all, V6 actually is almost perfect with the text, but there are some mistakes here you can see retro game with two A's. This is not great and sort of a pain to photoshop away. And 6.1 actually got everything right almost. There's just this double R here, which is less of a problem than on the last one. Again, the RR happens here and here is just an easier problem to fix, but it's not perfect, which I kind of hoped it would be. What about the Q2 quality setting though, if we change it to Q2? Yeah, we still have the double R issue here, we still have the RR issue here, but then some generations just do it here you see there's no RR issue and this was our first shot generating this. If I go back to V6, you can see that it doesn't even say retro, it just says retro reto. It just leaves out the R. And then one last comparison. If we use the style raw flag in the end you will see that the text actually works the best with this setting. The text accuracy also here is improved when comparing it to V6, but also here the M is not perfect, but it's close to perfect. So again, incremental improvement with midjourney. Plus there is also a new upscaler that I'm just going to briefly brush over. Aura SRV2 on file AI. They advertise this in a big way. Apparently V1 was a big success, but it didn't work for real world images. That's why they trained V2 now. And you can try it out for free on this hugging face space. We also went ahead and tested this and compared it to the midjourney upscaler. So this is the midjourney upscaler going from original image to the subtle upscale, which doubles the pixel amount and the creative upscale doubles the pixel amount. But make some creative decisions. You can see the shield is completely different. The background changes. That's what the creative upscaling does. Now, in contrast, if we look at the Aura SR V2 upscaler, you get four times upscaling. It is free, but there's a lot of sharpening going on rather than just upscaling. Upscaling. Sharpening has been a thing for decades now. We don't need sharpening, we want AI upscalers. But what we found this to work even better on is illustrations because there it really takes the lines, it highlights them well. So although this is meant for real world images, on illustrations is where it really shines. It does more just sharpening. As you can see on this video clip right here that we have from our testing. My overall verdict here would be that all of these upscalers that are coming out these days are pretty good. And it doesn't really matter if you use one over the other. There's some slight differences, but. But still, as I said when it came out, and I think it holds to date, Magnific, which is the most expensive option, is also the best option. If you're really looking to blow an image up and get high quality creative upscaling where the creativity actually works with the image, then I still think Magnific is the best. But if you just need a quick 2x upscaler, something like Aura SRV2 or just using the Midjourney Upscaler will work just fine. Oh, and if you made it this far into the video and you're enjoying it, consider leaving a like it really does help out not just the video, but the entire video, entire AI Advantage channel, and ultimately just helps us push things forward. It strengthens the channel, it strengthens the team, it strengthens the entire community around this channel. And we can keep producing high quality Gen AI videos for you twice a week. And now we move on to my favorite part of all of this week's research. All of this was very interesting. AI video just has a very special place in my heart. And Gen 3 actually shipped some features that I personally have been waiting for ever since the first Sora previews, namely its image to video. Plus they're releasing a second model which is even faster at generating Gen3 outputs and cheaper too, which is huge because Gen3 is really expensive. Generating a 10 second clip costs you $1 right now with the alpha version, but then it's really good. But how good is it and which one should you use? Well, instead of creating a standalone video, we did all the testing here for you. We ran various images through the top three models, Gen 3, Dream Machine and Cling and the result of the testing is actually a set of conclusions on which model to use when, which I'm going to present to you right now with the well tested Examp to back up my claims. So let me just state that all of these tools are top tier. I don't think there's a single winner. I think it's really hard to pick a single one that is best. I think arguably it is gen free, but different use cases require different tools and as we went through the testing of these tools, especially when we limited it to the image to video functionality, which by the way is one of the best things you can do with AI video because providing it with a lot of context in the prompt is almost impossible. A picture says more than a thousand words and you can provide a picture to to the model and then you can prompt on top, then you can animate it. It's way easier to craft a good picture and then making that move rather than trying to describe every intricate detail of your scene instead of a prompt to a video generator. And now all top models have this feature. So let's have a look at the differences. Let me just start out with the animation use case because this one was the most surprising. Okay, this is the Gen3 Alpha shot that stems from this image. Okay. Look at what it did. With a little additional prompting, we just added extra prompt to zoom out to this, and it did super well. Look at the character consistency when it comes to these little animated stories. It actually maintains that this is something that was unthinkable up until now. These little elves, or whatever they are, are actually consistent and recognizable across the entire scene. If you look really close, the only piece of artifacting is here on this present on the left. But I think that is quite acceptable considering that all of this looks wonderful and this is very cinematic. And that is one of my first conclusions. Before we even look at the other things. If you're looking for cinematic shots, if you're looking for dolly shots, big sweeping establishing shots of a drone orbiting around the subject, whatever it might be, Gen 3 is your friend. It's really good at these epic cinematic shots. This has just proven to be true across our multiple tests. But now let me show you the result of the same image put into Dream Machine. And this is what we got. Okay. It's just a slight animation, not much movement. And again, we prompted zoom out, showing the surroundings, but it actually didn't. It just did this. And that is fine because that is what Dream Machine is good at. It's good at animating things slightly. It doesn't overdo it. It doesn't go crazy. It doesn't artifact. That keeps things consistent. If you're working with logos, graphics, or product shoots, as we'll look at in the next example, Dream Machine is what you want to use over gen 3. Gen 3 is going to create artifacts and do these big epic moves and mess things up here and there, and it's going to cost even more. Dream Machine is really good for these slight animations. Okay. That's at least what me and the team found from the testing and our usage of these tools so far. And they've been out for a few weeks, and we kind of obsess over all of this. So I think across the team, we have enough experience to make claims like this. Okay? And here's the third one, Cling. And this is actually super interesting because, look, it actually made the mouse move. Now, this is a golden opportunity that I cannot pass up where I can just dub over them, right? Oh, if you haven't liked the video yet, this is your last chance. Go ahead. Sneaky, sneaky advertising right there. But I think this is wonderful. Cling actually went ahead and moved them. So very different results on the same type of image. And we prompted all of them to zoom out, but not all Did. Interesting. Let me go to the second one of three examples that I'll show off here. And that is this burger. Okay? And this is without a prompt. Okay? So look what Gen3 did. It just does this big zoom in, and it adds some extra lighting. Again, cinematic, right? This is big zoom. And look at the lens flare coming across. But in the context of a product shoot, this is not what I want. What I want is more like the result from Dream Machine, which you can see, see here. This is perfect. A light slider shot tracking the burger. And you have to realize it's actually coming up with what's on the side of the burger, right? It's generating all of this because it didn't see that in the original image. And it does it super well. There's no warping at all. Like, I looked at this 10 times. This is just a perfect product shot, isn't it? So Dream Machine, really good for these slight animations. And if you look at cling. Well, I would just say, generally speaking, it's just the crazy one out of the bunch. Sometimes you get really good things. That's why it went so viral on X, because some of the clips just are outrageously good, but usually there's just something that goes wrong. Something where you get a little bit of artifacts. Like, look at the sesame seeds on the top of this burger. They're just shifting ever so slightly. And you could never use this for commercial production. Plus, sure, the output quality is lower, but that changes if you subscribe to one of their plans, which, by the way, they have new plans now. You can purchase a membership now. They start at $5 per month, but at $46 per month, you get 8,000 credits. Remove all the watermarks and you can extend videos up to three minutes. But, yeah, these are the sale prices right now. Just check for yourself what you get. I have one more example that I would love to show off here to end on a bit of a low note, because none of these are perfect. Okay? This is an image of a woman standing on top of a sand dune. And let's just have a look at what Gen3 did with this. Okay? The woman is walking. There's a little bit of weirdness. This is my only issue with this, right? The walking is not perfect. But this shot, this drone shot, oh, it's so cinematic. You have to think about the fact that it invented this entire dune, right? It didn't have it in the original image. It just came up with the fact that, hey, this dune should be there and it should look like this. And it should be lit in this way with the gradient here. It's a beautifully composed shot. This is really good. But there is funkiness in the walking, so I guess it's sort of unusable. Let's look at what Luma Labs came up with. And yeah, typical Luma Labs fashion, the movement is way slighter. Now it goes for a little slider shot of the drone while tracking the subject. And I would even say the first second of this is usable where she stands as soon as this woman starts walking and starts being a little weird. But hey, not a single one of these free generators actually managed to do this perfectly because Kling just, I don't know, panned over and this woman looks all weird. This is sort of unusable. You determine which one is the winner for yourself. But I would just say it's just different tools for different use cases. Gen 3 is going to be epic and solid and cinematic. Waluma Labs is like hiring an animator that just makes things move and puts things into motion that weren't in motion before. And Kling is just a crazy friend that out of nowhere might just start consuming drugs at the brunch that you invited him to. Who let the dogs out? Who, who, who brought this guy along, then all of a sudden might start telling a story which is really fascinating, but most of the time he's just going to be annoying and out of place. So I don't know. That's my review of the top three video generators. And if you enjoy comparisons like this and you would love to meet other people that I have to say, same interests as you. That's exactly why I built the AI Advantage community and just want to highlight that. We did all of this testing for the community. There's a resource that compares it, there's discussions underneath. And now we opened up a brand new creative district which is all about video and image generation. And you can either participate in some of the events and to become better at these skills yourself, or you can just check out what other people are doing and learn from them. We also have a LLM and no code district. Now. AI space is evolving and so is the AI Advantage community. So if you have some time in August to learn, honestly, the monthly prices are still considering everything that you get in there. You could easily spend all August, eight hours a day just learning about Gen AI tools with all the materials in the community. I just wanted to highlight that I've been told that I don't do that enough. So there you go. The community is the ultimate educational resource, the best thing we've ever built. And I think at the monthly price, it's quite accessible, considering what's inside. And that's all the news you can use for this week. I'll see you next Friday when we do this again.