They Just Released An Image Generation Tool Thats Too Realistic...

So there is a new AI image generator that is even better than the prior state of the art image generation model, Flux Point 1, which is now Flux 1.1 Pro, the faster and better model. This model is absolutely outstanding as this provides six times faster generation than its predecessor, Flux Point 1 Pro, while also improving image quality, prompt adherence and diversity. And you guys really have to understand the implications for Flux 1.1, because this company has seemingly come out of nowhere and has just stunned the AI industry with all of its amazing capabilities. I think this was one of the things that really surprised everyone because when we looked at who the state of the art companies were, we knew that it was currently Midjourney, Ideagram and Dall E 3.5. But a company coming out of nowhere was rather surprising. Now, of course, this is actually the model that I thought was the secret OpenAI model, because it was called Blubri and I thought this was OpenAI dropping their image generation model that was a part of the SORA project and just kind of linking it to the Strawberry model. You can see here, it says Flux 1.1 Pro and has been introduced and tested under the code name blueberry in the AI image analysis arena. And it surpasses all other models on the leaderboard. So this isn't a model that they've come out and you know, they're just like, okay, we're just going to bring out a model. This model currently surpasses everything on those leaderboards, which is why I said, this is incredible. And this thing is going to be hooked up to so many different companies before you know it. So let's take a look at some of the things here and let's see exactly what's going on. So we have this chart right here, and this is of course the Elo score for different AI models. What we can see here is that Flux Point 1 Pro just lags behind Flux 1.1 Pro. But you can also see that it clearly surpasses Ideogram V2, Ideogram V2 Turbo, and Mid Journey version 6.1. I think one of the key reasons that Flux 1.1 Pro manages to surpass these other models is the Fact that Flux 1.1 Pro actually listens to exactly what you want to say. This is what you call in the AI space, prompt adherence. This is basically where if you try to get something really specific, let's say you're imagining a frog sitting on a bottle of ketchup that is on top of the moon. Flux 1.1 Pro is quite likely to do that. Whereas images like Midjourney, they try to make the image as aesthetic as possible rather than actually sticking to what's in the image. That's why Flux 1.1 Pro has actually won out on many different benchmarks, even if some of the images don't look super photorealistic. So let's take a look at this because I think one of the key things that most people don't understand about this model and, and why it was such a huge thing when it was announced was because of how those images actually look on a day to day basis. So what we can see here is a Reddit post that went absolutely viral in the ChatGPT thread where users were basically saying that it's over. These are all AI. Now I'm going to flick through some of these images and we have seen images that look super realistic before, but I'm pretty sure that Flux's training dataset largely contained raw iPhone photos, which is why the images that it generates can be so surreal and photorealistic with pretty much no way for you to tell that it's AI generated. So of course we have a woman here. I think this one does look good, but there is a small issue with the mouth that I can see that makes it slightly unbelievable. But let's not focus on that image right there. When we can look at the other ones that are far worse, we can look at this image which doesn't really have any AI generatedness to it, if that even makes sense. It just looks like a raw iPhone picture. And when I saw this, I kind of started to, I wouldn't say worry, but be a little bit concerned for what is now reality, because this picture doesn't have anything that suggests that it is AI. One of the biggest problems with nature is that there is a sheer randomness to it, meaning that we don't expect certain amount of shapes or sizes for certain rocks if they are completely random. In fact, it looks even more realistic for it to do so. So pictures of nature, I mean, it's not like any harm can be done done with them, but it certainly doesn't help you trying to distinguish those from ones that aren't real. We have another picture here of a walkway and once again, this just has that iPhone raw feel to it where you can't really tell. Now I'm going to show you a bit more and as we get on into the video, you're going to see that the further on we get, the more realistic these pictures start to look. And the craziest thing about this is that while I was trying to research certain things for this video, I. I tried to find certain images that were real. But the problem is, is that there were so many images that were AI, you know, generated that I couldn't even find enough to get a comparison with. So for those of you who talk about how AI may have polluted the Internet, that argument is starting to become more and more apparent as we get on. So you can see right here, this is another image that looks shockingly realistic. Once again, another image here that also looks quite realistic. Until we decide to, I guess you do further analysis upon the face. It doesn't look too realistic if you look too closely at the face. I mean, it doesn't look fake either. But this picture, I really like this one due to the lighting. It just has that. I'm not sure what you call it, but it just has imperfections that Midjourney and other AI models don't have, which makes me believe that it is realistic. There's something about the imperfections of, you know, poorly taken photos that just make them that much more believable. And then we have this one right here that I would literally have never guessed that this is AI generated at all. But of course, as you know, it is AI generated. Now I'm going to dive into some more images and you're going to start to probably lose your minds at how much this software has progressed. So we have this right here. And this is a meme, of course, before we get into the worrying stuff. You see less AI generated images. It's like a happy face. And it's like, wait, I'm seeing less AI generated images. Which means that the images that I am seeing, I don't know which is AI generated or not. And I think this is going to become more apparent as the future takes place. Although I do think that in the future there's going to be a lot more tools in place that allow people to identify if images are AI generated. Such as potentially when you make an open source AI tool, it must have some kind of, you know, watermark that is extremely hard to remove when doing image generation. For example, if you do use Instagram, what they will have is a watermark that tells individuals that this picture is AI generated, which is a good thing because misinformation is pretty bad. And this was one of the images that I found on Twitter. I'll leave a link to it. And this is a lot better than the images I saw last time, which with Flux's realism, it was completely shocking when I saw this Because I was like, I'm looking at a lot of small details, and there isn't one small detail that gives this away that this image is AI generated. I would argue right now that even though someone has told me that this image is AI generated, that I almost don't believe them. Like, I still would like to see the origins of this picture because it looks too realistic. Like, there isn't any subtle mistake that I can see. And maybe you guys can see one. I don't know, maybe you guys can see one. I can't see one. But it just has that, you know, I guess you could say awkward lighting and awkward pose. As if they just took this rather quickly. And his face is like, okay, I'm in the picture. But she's clearly excited and I'm not. You know, it just has that weird vibe that just makes this seem super realistic. So my brain is struggling to truly understand that this is. Is AI generated. But there are some images in which you can tell that this is AI generated. For example, you have this one right here. And upon first glance, you're thinking, okay, that might be AI generated, but, you know, all that might be realistic. But if we do take a look and then we look at this bottle and we look at the labels, we can see that it is unrecognizable symbols, which allows us to understand and identify that this is completely AI generated. Now, I'm not sure exactly what is being served on the plate. It does look like corn, maybe some onions, I'm not entirely sure. But those are largely the biggest things that can help us identify whether an image is AI generated or not. Currently, these systems do struggle with generating text, but as the old saying does go in AI, this is the worst it will ever be. Now, there are some tips and tricks for those of you who want to try this software out. If you are going to be trying out Flux 1.1 Pro, one of the prompts that allow you to get this hyper realism is using the Image HEC extension. You know, if you prompt as image hec, which is the common file extension default on iPhone photos, it will result in super realistic pictures. What you can also do is you can also add that in the first bit and then add dog or add cat, and that will allow you to also specify what you want in the image. Now, I did try a few images myself, and I have to say that even though I've generated them myself, I'm still struggling to believe that they're AI generated. For example, there was this image that I thought looked A bit too realistic. Until I realized that upon closer inspection, we can see that this area does look AI generated and looks like a bunch of mess, which is good. And also this image, whilst it looks super realistic, it doesn't have any mirror. So this picture doesn't really make any sense whatsoever. Also, I did manage to generate this image. I did, I did manage to generate this image, which does look really realistic. But if you do zoom into some of the key details, you'll start to see where the errors start to arise. For example, if we take a look at the license plate down here, you can see that that doesn't look like any sort of license plate. And the linings around the car sort of merge and mesh together with the house as we start to look at this even more. Now, of course, certain images can be upscaled and this image was upscaled from the default output, so this was upscaled from freepik. But you can see as we start to dive into the real core details of this, we can see that the AI generation just kind of, you know, messily puts things together and I guess maybe in future it's going to look a lot more better upon further inspection. But the thing is, is that when you look at what's going on here, the average person doesn't have time to look at all of these subtle details and really analyze what's going on. Oftentimes when things go viral, it is just like a headline and an image rather than people diving into what's going on. So that's why this stuff is super important. But like I said, you can see certain imperfections in objects that are supposed to be certain sizes or angles, which is why half the image looks really weird. But then if we go to the nature side of the image, this looks a little bit more natural because like I said before, nature has these unnatural shapes and the randomness is inherent in their original design. So that's what makes it super believable. Now, this was the last image that I generated and once again, if you zoom in on further details, you can see that it does look pretty nice at the first glance. But once again, if you zoom in, you can start to see all of these messy issues. Now, no software is completely fail safe. There's going to be issues here and there, there's going to be mis generations and miscalculations and some images, you know, people have three hands, they have four hands. But overall, I still think this is a remarkable piece of technology that people should be aware of because now many individuals are going to be questioning what is reality from what isn't. Let me know what you guys think about this, and I will see you in the next video.