Discover Midjourney V5's Hidden Powers with GPT-4!

Another massive release just happened. Midjourney V5 just came out, and I already found some tricks to get better results. But why would you even want to use this? And once you use it, what do you need to look out for? We're going to be discussing all that, plus how to use GPT4 to get superior midjourney v5 results today. So let's get into it. So, first up, I need to tell you that only paying Midjourney members have access to this. But if you're a paying subscriber starting at $10 a month, it's quite, quite easy. You just go into discord, say settings, and now you can change your model. This is the most comfortable way to interact with it. Otherwise, you need to include V space 5 at the end of your prompts. But why would you want to use it? Well, simply put, the biggest reason to use v5 against v4 right now is a desire for more realistic images. Concretely, humans, they finally fixed the hand situation, so sometimes you still get an extra finger, but it's not a catastrophe every single time anymore. Also, the photorealistic qualities of this thing just went through the roof. These pictures look as if they were taken with a real camera, like they're 95% there, with great skin texture and proper hands and even details like teeth now being properly represented because they trained this new model to do exactly that. You can start using this thing to replace models right now, but the more realistic settings that are baked into the system are not everything here. Overall, you can view V5 as more of a Pro model than V4 was. You need to go more into detail to get better, better results. So, as opposed to GPT4, where it got smarter and it got a little easier to prompt, this moved in the other direction. You can get better results, but your prompts need to be better. And here's one of the biggest changes. No more prompting with keywords only. At least they don't recommend it. You should use natural language, just like I'm speaking to you in right now, just like you would communicate with a human. If you just take a step back and think about this for a second, you'll realize that natural language includes so much detail and so much direction that obviously it's going to be able to craft better results from it than language, including details better. All those little words in between actually do a lot to communicate what we want to communicate. So no more keywords only. But once I show you the prompt that I crafted in GPT4, you'll still be able to use keywords because GPT4 is going to do the natural language processing for you. But more to that soon, another change that I personally really enjoy is that it's upscaled already. So no more waiting twice if you want to upscale something. So if I just generate a cat with a hat right here, right away, you should be able to see how realistic these are by default without me including anything like photorealistic or aperture. All I asked for is a cat with a hat and I got photos of a cat with a hat. So yeah, I just realized these photorealistic capabilities are baked in now. And as I mentioned, the upscaling is super fast. So in real time I'll just click U1, meaning upscale one. And without even finishing my sentence, the thing is done. This looks way too real. If you would show this to a normal person. I think this passes a photo as long as a person is not consciously looking for AI generated features. With my photography background, I would say the depth of field gives it away a little bit. These whiskers should have been sharp because the ones on the left and the eyes are sharp too, and they're in between those two focal planes. But that's just me nerding out. Now let's move on to the interesting part and that is how to get better results by using GPT4 for this. So what I did here is I prepared a little prompt for you guys. So what you want to do is you want to go to GPT4 as the results you get for this are slightly better than 3.5, because we're kind of pushing it here with the amount of detail we request. And you will be able to simply copy paste this from the description of this video. But essentially it makes GPT4 act as a stable diffusion photography prompt generator. And then we specify the input and the outputs. So we input the visual description and it outputs a detailed paragraph that I can copy into my diffusion model. Include a variety of photography related terminology, including the description of the exact Zeiss lens. So I specified a brand here because that's my favorite look. And that's also a lot of the equipment I use around here because I just love that sharp look. It looks very professional and very premium, very luxurious. And this is what I want for my images too. But you can switch this up to any brand. You could even say iPhone lens to imitate photos a normal person would take in their everyday life. Either way, the prompt continues, and most importantly, a detailed description of the volumetric lighting. So if you didn't know volumetric lighting is a term that I borrowed from my background in videography and cinematography. And it's what some people refer to as God rays, Essentially, this diffused beam of light in the background of an image. Again, you can change this for a different type of lighting, but I really enjoy this in my images, as in combination with the Zeiss lens makes it epic and luxurious, something I really enjoy in my images. And then we end with. Now write me a clear and concise natural language visual description of the following. The natural language part here is very important because, as I said, it doesn't want keywords anymore. But the cool thing is I could go ahead and use keywords to describe this now, because I have a prompt generator. So let's do this. And my idea here is let's generate some prompts that recreate items in my room. For example, there's this ship inside of a bottle right on my table. So let's try and recreate the epic version of this inside of Midjourney V5. Okay, so all I'll do here in the end is say ship inside of a bottle and say enter. That simple. And what it does is create a detailed description of my scene, including a specific lens, and I'll squeeze in this little tip. The lens here has become increasingly important as these outputs are very realistic. Just like in photography, One of the main things that influences that look is your lens. So there never has been a better time for you to learn a little bit about photography. And although I would love to sit here for a few hours and give you a basic lecture, because still I'm so passionate about all this camera stuff, I'll make it simple for you. And I included two resources in the description that will give you a basic overview of lenses. Okay, so look over here. Sigma is one of the main manufacturers of lenses and they have this fantastic page which shows off different focal lengths. And focal lengths are essentially how wide your shot is. That's all you need to know. And whenever you want a specific look, you can simply look at this page and pick this millimeter number right here and switch that out inside of your prompt. Now, if you want to get even more advanced, you could pick specific lenses. And a good way to do it is by going to the website of b and h photovideo.com and when you go to photography and lenses, you can pick from all the different brands. So just pick SLR lenses because there's the most variety there. And then we'll do something most people won't do. List from high to low. We want the most expensive ones. Right. And right here you already see that it actually shows the Zeiss otus lenses, which GPT4 picked by default. I mean, how smart is that? And the OTUS lenses are very well known for portrait photography by many people. They're considered the holy grail of portrait photography because they're super expensive, super sharp, and GPT4 already uses them. But you could just look at this bundle and here in the description, you have all the lenses in the bundle. So I could just copy this 100 millimeter. And again here I would just check. Ah, 100 millimeter is really zoomed in. Yes, that's kind of what I want with my ship inside of a bottle. Right. And once I copy my prompt into midjourney again, I'll just say, imagine, as per usual, I could just go ahead and copy the name of this 100 millimeter lens. And that's the only thing I'm going to change about my prompt. Okay, and now I'm going to hit enter. So now it should be less wide and more focused on the details, which is exactly what I want with something as detailed as a ship inside of a bottle. Now all we have to do is wait for a minute and look at the results. Whoa, whoa, whoa. Look at that. I mean, that's pretty impressive. Look at that. That looks so realistic. This could easily be a real photo. Now, given I could have included a lot of specific details to make it look just like this, but honestly, I just wanted to see what midjourney plus GPT4 would come up with. Okay, to round this out, I just want to do one more because I know in V4, this prompt always generated an image of something very cartoonish looking. I'll just say AI youtuber at his desk. And funnily enough, it actually uses a lens that I typically have on top of this camera, the size bis, but I usually have the 40 millimeter. It really went hard on this, this description. So let's see what we get here. And I'll just switch this to the 40 millimeter and hit enter. Oh, wow, look at this. Okay, given there's many details which are messed up, like what is like my journey, develop some camera prototype that I've never seen before. But look at this guy sitting at his desk. And the hands are not good, but they're also not terrible. They used to be just terrible. Right? Okay, so that's it. If you create some amazing images yourself, I want you to join our Discord server and share them with the community, me and many others in there are posting their creations and discussing what is going on in the AI space on a daily basis. And if you want to get even better at midjourney, you should check out this video because it will teach you the ultimate hack on how to get the most out of it with just one keyword. See you there.