Unbelievable Week for Realistic AI & More AI Use Cases

Okay, ever since summer started, this is insanity. We have been accelerating with AI releases like, I don't know, I guess April 23rd had the same frequency with GPT4, etc. But let me tell you, this week was wild and particularly in the creative department. We got so many state of the art tools features both on the open and closed source side. There's really a lot to look at and we went ahead and researched and tested it all for you. Just to summarize, all the results in this video, ranging from two new state of the art image generators, both from Google and Black Forest Labs, all the way to every video generator releasing something new this week with one feature really being a standout. Minimax just seems to be better than anything we've seen so far. More on that later. And then also a little preview at what Meta is doing with their video generator that they just teased. But we took some of the prompts they showed us and compared those results to some of the best today. So let's get right into it in this week's episode of AI News that you can use. Okay, so today we're gonna start by talking about Meta's movie gen. This came out last Friday, so we barely missed it in last week's episode. But don't worry, we're going to have a look at it today and this announcement by Meta is packed with things to discuss here. Now, rather than looking at and criticizing or complimenting every single thing that they show us on this page, you can do that yourself. And I today I'm going to take a bit of a different approach here because what I had the team do is take some of the prompts that they presented us here and we tested them in two other top tier video generators. Now, yes, this thing can do a whole lot more than just video generation, can also do video editing where it alters subjects, and it seems to be top tier at all of this. By the way, when I say top tier, I'm sort of referring to this monthly ranking that we update in the free area of our community. You can check back in at the beginning of every month. And we update these for video generators, image generators and LLMs, and the top tier ones are these four. So although MetaMovieGen is not available just yet, we can already compare it and have a look if this will be worth our time once it's out. Okay, starting off with this ghost in a white bed sheet facing a mirror. Obviously a tricky shot due to the mirror, and we'll also keep an eye on the shadows here on the left that it produces really well. Now, mind you, we didn't take image references here. We just ran the prompt through the generator. So this is going to look wildly different, but we can't judge the quality. All right, so here's the result from Runway, which is a whole lot more abstract. And the whole mirror thing didn't exactly work out. I mean, I guess there's a mirror element to it, but, hey, this is not even close, is it? Let's have a look at the same thing inside of Minimax and see about that. Okay, so this looks closer to the original. Look at that. The way it mirrors. Not fully perfect, but quite good. And the shadow play here on the left. Yeah, this moves along with the Ghost. I think this one is actually closest them. As you'll see later in this video, Minimax is completely killing it, even with the new features that came out. More on that later. But next up, we gotta take a look at this little monkey with a boat. Okay, so obviously this looks really good. I think in the beginning, you can spot slight inconsistencies in the face, but other than that, all the fog, the physics on the boat are. I don't know, saying indistinguishable is maybe a stretch, but if you're not concentrating on it, you would never catch the fact that this is AI. Okay, this time, let's start with the Minimax version of this. And what is this? Is a bit oversaturated. The physics look. Okay. The fur looks good. I just don't like the quality it has to it. That truly could be prompted out or just reran. This definitely looks more artificial than the meta one. And here is gen 3. And again, this is not even close. Okay, one last one. This little sloth with pink sunglasses laying on a donut float in a pool. Nice little clip. Let's see what Runway Gen 3 did with this. Oh, my God. What the heck is going on over here? That's not a Sloth. That's okay. Runway completely failed this test, in my opinion. And Minimax. Ah, there you go. That's sort of doing the job. I mean, if you want to do a social media clip with a sloth chilling out inside of a float, that does the trick. Yes. So, wow. For next month's ranking upgrade, it looks like we're gonna have to move MiniMax ahead of Gen3 here. And then once Metamoviegen comes out, it seems like it will take rank one. One more thing that I want to highlight here is the ability of the model to do one shot custom videos, meaning you just give it one image and it creates custom videos of you like this. If you want to see more examples, you can just check out the page. They did a great job of showcasing this all. So although this is not quite available yet, I did want to feature it because as you can see, Minimax that is available today produces comparable results. I think the best thing about my job here is that every week I get to learn new things. I mean, AI has been moving so fast ever since summer ended that I feel like I'm learning a new thing every single day. And I'm so grateful that I have this opportunity because I've always been super passionate about leveling my skills up, which I guess makes me a true believer in lifelong learning. And if you share this passion, you might want to check out the sponsor of today's video, Brilliant.org Brilliant is an online learning platform with hundreds of courses across math, science, data analysis and more. My favorite thing about Brilliant is how hands on everything is. I was never a fan of sitting in a classroom and listening to a two hour long lecture on something that barely interests me. I've always learned better if I could actively engage with the topic. And that's exactly how Brilliant courses are designed. And if the hundreds of courses they have on the platform feel overwhelming, well, no worries. Brilliant has these learning paths that organize them into clear paths that you can take. I particularly like the computer science and programming learning path here. Finishing it gives you a great foundation of programming skills which can only be helpful when interacting with all these tools that might not require code code, but they're still technical. And if you ever wanted to build something with AI, this would be a really great place to start. So go check out Brilliant. Using the link in the description. You can try it out for free for a full 30 days. And if you decide to stick with it, you'll get 20% off the annual subscription. And now let's move on to the next piece of AI news that you can use. Okay, so now let's talk about two significant releases in the AI imaging space that came out over the course of the last week. The first One is Flux 1.1 Pro, a slight upgrade to to Black Forest Labs, the open source and Hyper Realism King. And the second model that came out is Gemini's Imagen Free. Now, if you're not aware yet, Imagen is Google's flagship image generator. Maybe not quite as good as Flux or Midjourney that in my mind share the first spot, but they have consistent text, they have Hyper Realism they have prompt consistency, they have really good prompt adherence, and it comes with your Gemini subscription, if you do have one. So what we're going to do here is we're going to do a live shootout. Okay. I'm going to take three random prompts from our big resource that we created over a year ago and that now is updated and available to our community members. But basically, this compares midjourney results dating all the way back to Midjourney 1 on a total of 100 prompts. So what we're going to do here is take three of these, run them through Imagen, run them through the new flux model, and then I'll fully subjectively pick which one I like better. And then here, you can obviously compare it to the newest midjourney versions, too. Whatever model gets the most points is going to be the winner for this week. And then in November, when we update our public rankings, we're going to do more in depth testing. So let's begin. Okay, first prompt here. How about this lioness? This looks fantastic. So this is going to be an example of a hyper realistic wildlife photograph. Starting off inside of Gemini Advanced and Imagen free. I just removed the image prompt in the beginning and the aspect ratio here, and let's have a look at what we get over here. I have Black forest Labs Flux 1.1 Pro. I'll run this one and let's have a look. Okay, the first results are in. Let's review. Okay. Whoa. I can already tell this is not going to be easy. Let's put them side by side. This is almost impossible to judge, to be honest. I mean, if I had to be super critical, and this is a personal pet peeve of mine, the focus on this image is clearly on the lioness's eye. Right. But then also, these whiskers are perfectly in focus and this eye is out of focus, that's just physically impossible. So this is usually a telltale sign for AI image generation, as the way this works in real cameras and real lenses is that there's something like a focal plane. So right now my eyes are sharp, but the background isn't. Right? Right. But if I do this with my hand, you will see that now it's also blurry. Right. And as it gets closer to my face and enters the focal plane of my eye, now it's sharp and there it disappears. So as you can see in this flux output, one ear is blurry and the other one is perfectly sharp, which is impossible because they're supposed to be in the same plane. And this Issue is not present on Imagens. So first round, I would give the image in here, but also, let's have a look at the Mid Journey generation. That's really good still. Okay, you know what? We're gonna take the midjourney images as reference and make this a battle between the two new models. But I would say Midjourney is on par with Imagen3 on this one. Okay, next up, let's take a completely different image. How about taking something that has a level of copyrights to it? Right? This would be interesting. This Banksy style image should be a great example. Let's see how lobotomized these models are. Again, remove the aspect ratio and the imagine part for midjourney, run this in Flux, create a new chat, Run this in Gemini Advanced. Hmm. And on this one, let me tell you, I've seen various Banksy exhibitions, and I'm actually a big fan of the guy's art. I don't know if I would identify these two pieces as a Banksy style artwork. Whereas Midjourney really nails this. Right, let's give him one more chance. Okay. Consumerism sent this looks a lot more like Banksy to me. There's the message. There's a sad child inside of a shopping cart, and the font matches up. Okay, solid. What about Gemini Advance? I'm a bit afraid of this generation because obviously Google always censors their stuff. But yeah, I mean, to be fair, this does look a lot like Banksy. But what about the message? You're in love and as Coley low to. So it didn't do a good job of doing that at all. Maybe I should have specified what text I concretely want from it. But, hey, Flux made this happen midjourney. Definitely made this happen without the text because it can't do it. So if this one is between Flux and Imogen 3, I give this round to Flux. Now let's move on to the final round, which we'll do on something tricky, shall we? Okay, how about this anthropomorphic creature, meaning a human, like, fantasy creature. Flux 1.1 over here, run. Google's Imagen 3. Over here, run. So we're going to be looking at anthropomorphic majestic blobfish knight. Ooh, and here we have the Flux results. This is great. This is definitely a blobfish knight. Okay, so Flux is clearly a little faster, and obviously this is open source, so people can build it into their own apps, chatbots, whatever. Whereas Imagen is constrained here. But who? Okay, so there's differences. This One is closer up, but this one is higher detail, isn't it? Man, this is the first image that makes me want to download this and really zoom in. Look at that detail here. This is before any upscaling. Wow, this model is really excellent. Look at the Mohawk, the golden fish Mohawk. How sick is that, honestly? Okay, so as you can see, I'm obviously a big fan of this picture. And how about this one? Let's also download it to make this fair. Hey, I gotta say, the quality on this one is really good tool. If I go into the details, like the eyes, I might even prefer this one over here. But hey, the golden Mohawk. Mohawk, come on, that one's hard to beat. Plus the armor, I mean, this one is a bit subjective, but I do like this armor a whole lot more. I kind of want to give it a tie. But hey, if I had to pick a winner, I would say that Flux is a bit better here. But let's be real, these are both really good. So Flux 1.1 pro winning two. One here. A final look at what Midjourney did here. Also excellent. A whole lot more detail in this one, right? Look guys, these image generators are just so good at this point. If you pick one of these S tier options, with now Imagen 3 being the third one up there, you're gonna incredible results and I hope this little segment helped you in deciding what you want to use for your next project. If you have been following the rankings that we're doing in the public area of the community, which is completely free, you don't even need an account to access this. We have these different rankings for various AI tools that me and the team put together and update once a month. And for the October update of the video ranking, Minimax, this new company that we also introduced here in the show a few weeks ago is actually in second space. We find it to be better than most of the competition here. And they added the most crucial feature this week, which is this right here, which I can't really read. So I'm going to go ahead and translate this to English. Upload. Yeah, there you go. So as you can imagine here I can upload all sorts of file formats and then it will use that as context to generate the videos. And here's the thing, this is the best image to video in the market. If you're looking to turn images into something that moves. Minimax by Hailuoai is what you want to be using. And I'm not just saying that. I can prove it. Okay. A few weeks ago, when Runway and Dream Machine introduced image to video. We tested it and compared it across a few images. Now, we ran the same images through Minimax here. And let me tell you, during our research and testing meeting with the team, we were stunned by how good this is. I'm not even kidding. Just have a look. Here's what it did to the simple animation. And all of this have been processed without a prompt, just like we did in the others. Now, for comparison, here's the result from Dream Machine. And here's the result from Gen 3. Minimax is just the best. I don't think there's any arguing there. I have a few more examples here. We ran it for this one. It's okay. It's not what I would expect. I wouldn't say it's the best in this case. But when it comes to this woman walking across the dune, all of the other generators failed miserably. And also Minimax sort of failed here. But I'll say this especially. The first second is sort of usable here in the middle. Somewhere at second 2.5, it starts falling apart. Fair enough. But then the end is good again. Meaning if you would regenerate a few times, you would actually get something usable out of this. Whereas the other video generators all just had useless outputs. For example, animating this image, it did super well. Subtle movements. Balloons are moving around independently. You can see that, especially on this one on the top left. And I really like the subtle animation here on this one. It's on par with some of the others. And I would say on this burger, it's arguably a bit worse than Dream Machine. Because Dream Machine really nailed this shot here. The burger is spinning, which is not what you want in a product shot like this. Maybe you could prompt around that. But we're judging this at face value. And here's my actual favorite one. This car I had high hopes for with the other video generators. But it didn't work out at all. It sort of just sat there and the background was moving by. So it actually animated it in a meaningful way. The background movement looks perfect. I really like the motion blur and some of the touches it added there. But most importantly, the car moves through the frame. Sure. The tires aren't moving. That's the one thing that I could nitpick here. But other than that, this sequence, if you need a shot of a car speeding through the frame, this is the totally usable in some sort of production, if this is the style you're going for. And then we have this man here, which is very interesting. I Don't know. I'll leave this one up for you to judge compared to other generators out there. But overall, on the shots that matter, animation, lifelike narrative scenes, and this sort of commercial shot of a car, I would argue these are the most important ones of the batch. And on all of these, Mini Max did perform the best. Do you disagree on any one of these? Leave a comment below. We can talk about it. So to round up this AI video segment, I want to show you one little workflow or tool that can kind of bring all of this together. It's really interesting. I wasn't aware of this MERFAL AI. We talked about them before built this web based comfyui that you don't have to install. If all those words don't make any sense to you, don't worry. Let me just show you what this can do. And basically here you have one input where you can add a prompt. So in this case a fat cat with an AI hat. Okay, great. That's our default prompt here. I run that into a Flux Realism Lora and hey, this workflow right here I built in like two minutes maybe. So Flux with a Realism Lora generates this image of our cat. By the way, I'm loving this image. Looks so good. And then and here it starts looking tricky, but I just send it to three different nodes. One of them for Luma Dream Machine Image to Video, one of them for Cling Image to Video and one of them for Runway gen 3 alpha image to video. Now we don't have Minimax here yet because that just released. It's just a question of time until they release their API and we will have that in here too. But this is a great way where you can enter one prompt. So let me just modify this and just say a cute cat with a hat. I run this and as you can see can see it starts generating the image and as soon as that is done, it feeds the image to all three of these. So if you really care about your video generations and you have the budget to do this because yeah, you do need money to run these workflows. This is the API. You pay per request. Then this is a fantastic no code way without any installations, nothing else except of an account with file required to test different models or build simple little pipelines like this where you input a prompt and you get four results instead of one. All right, there we go. It took a bit, but now we have three cats with hats and you can just compare them side by side like so. And I think in this case Cling sort of wins. Okay, good. I just wanted to show you this is really easily possible with a few dollars in credits. And now let's move on to the next one, which is a new quality standard in AI avatars, meaning AI generated versions of you. And this is a new feature for Hadrian Avatars 3.0, which we looked at a few weeks ago when this came out. Remember, we created this little avatar of me. It's pretty good. I guess in the recording, I should have looked at the screen a little less. And what you can do is you can head on over to your avatars. And by the way, I'm on the free plan here, right? So you can try this for free. Obviously, if you want to generate more than free videos, you need a paid plan. But now if I go on here, I can add an additional 99 looks. Okay, so up until now, you had your avatar and you could train a separate avatar, but now they have this looks feature. What does this mean? Well, this means that if the avatar changes location, like in this demo video here, she's sitting on the couch, and then here, all of a sudden, she's leaning onto the couch. You could do that. Look, now she just moves forward, changes outfits. This really sells the video. Even if you check out this YouTube video, the way I present this now, I'm sitting upright and presenting, but maybe in a sec I lean onto my table a little bit, I do something with the screen. This is a different setup than this. Right now you can train both of those into hey Gen and then edit it together and get superior results. Even doing multiple outfits and moving around the scene. I just wanted to highlight how useful that is, because if you're not aware, my background is in video production, namely educational video production. And even there, adding a little bit of motion, changing the outfits, the cadence of the speaker, all of that adds to the experience. It's not just TikTok that demands it in all forms of media. You kind of have to switch it up a little bit. And this is a really welcome feature that does that. Now, at the very least, I would recommend you check out a few seconds of this video so you're aware of how far these tools have come. It's not perfect, but it's quite good. And yeah, these things are also progressing, just like everything else in AI is. I can't wait to see how long it will take before I can actually fully replace myself in a segment of the video and nobody will notice. I'll definitely try at some point. I don't think we're at that point in time yet. So those are the releases for this week. But if you're new here and you want some actionable advice on how to use AI, make sure to sign up to the AI Advantage newsletter. On sign up, you get a massive template with a lot of presets, and once a week we send out one roundup email, catching up with all things AI in no time. Other than that, feel free to like and subscribe if you enjoyed this and I will see you next week.