Amazing NEW Features - Midjourney 5.2 Deep Dive

Midjourney just released their version 5.2 with amazing new features. This takes AI image generation to the whole new level. The most important one is the new zoom out feature. The next one is the new shorten command that lets you analyze your prompt and make suggestion on which words probably are not contributing towards the image generation. There are changes to the overall aesthetic system as well as there is a now a new high variation mode. We're going to be looking at all of these in this video. In order to activate this new version, simply go to Settings and then in Settings you need to select the MJ version 5.2. This is the default version now. Okay, so let's look at a very simple prompt. A close up image of a sports car in a busy screen. Now pay attention to this close up word. Okay, so here are our images. And these are pretty nice looking images. Midjourney is great at creating some amazing images. Now we need to pick one of them. So I'm going to pick this image. Let's say I'm going to pick the first image and we'll simply upscale that image. Here's the upscaled version. It looks pretty great. Now if you go down, you're going to see some new options. So for example, there's custom Zoom zoom out 1.5x and 2x and then there are two different variation options. We're going to look at all of them. But first let's check out the zoom out 1.5. Okay, so here's the zoomed out version and it's simply amazing because it added a lot of details in the background. So here is the original image and here is one of the zoomed out image. Now if you pay close attention to it, it added some cars in the background as well. And in the original image this person was not fully visible. In the original image, the person is partially visible as you can see here. However, in the zoomed out image you can actually see the person. Now here's the trick. You can actually zoom out it further. So let's say we're going to use the 2x zoom in this case. So here's the further zoomed out version. Now the great thing about this is you can keep doing zoom out on the images and it will add a lot more details in the background. Now essentially this is like outpainting that stable diffusion or Dali desk. However, it's not adding for further pixels in the image. So if you zoom out by the twice amount, it's not going to add twice of the pixels. The number of pixels is going to remain the same. So you will lose some detail on the car. But still it's an amazing feature to have. Now the next option I want to show you is custom zoom. And this makes square. In this case, I generated an image with the aspect ratio of 16,9. Now using this mix square option, it will convert it to a square, but it does something interesting. It actually took a 16 by 9 aspect ratio image and added details in the top and the bottom. Now this is a great way if you want to center your object. Now let's go back to our original image. And here you can not only change your prompt, but also define the custom zoom. So let's take it to a crazy value of 10. So hit submit and let's wait for it. Okay, so it seems like it cannot be more than two. Now, in order to show you what these other two options do that were very strong and very subtle. So I created another image. So realistic headphones with smoke coming out of the earpieces, DSLR ultra wide. And then we are using an aspect ratio of 16 by 9. Right. So first let's check out this option. So very subtle. Now I'm in the remix remix mode. That's why it's asking me if I want to modify the prompt, but I do not. So let's simply submit. Now, as a result, it created four different images, but they are very similar, both not only in style to the original image, but to each other as well. Now let's see what happens if when we try to vary it by a stronger scale. So again, I'm going to just submit. Okay, if you now look at this one, there are more variations across these four images compared to the previous run. Now this is where we set the vary to subtle. And you see the orientation of all the headphones is very similar along with the smoke coming out of it. But in the second case, the orientation of the headphones in each image is relatively different compared to the rest. Now, if you like to write very detailed prompts for image generation, this next one is going to be a game changer for you. So there is a new option shortened where you can simply type in your prompt. So I'm typing this prompt. Muscle car, aggressive styling, racing stripes. It's a relatively big prompt. Now after running the shorten command on it. Now, if you look at the results, you will notice that in the original prompt, some of the keywords are actually highlighted or bold and some of them are just striked out. So those are the keywords that are not really contributing anything to the image generation process. Now you can go and look at the further details. So if you click on show details, then it actually shows you the weight that it's assigning to different keywords. Now this is just amazing because it will help you in writing really good prompt. So essentially Midgen is doing prompt engineering for you. Now there are five different variations depending on the weight of the keywords. So what I'm going to do is I'm going to create images based on these five different combinations. So in order to do that, simply click on the numbers. So let's click on number one. Okay, so here are the results based on these compact prompts. And actually it does preserve the intention that I had. So it's muscle, current stripes, blur, Action shot, teleporter. And even for the just like a four or five keyword prompts, they still do pretty good job. This shortened command will change the way you prompt because you're going to be including only the most important keyword from now on. So make sure that you make good use of it. Okay, so the last change I want to show is this improved stylus command. Now it can have a value between 0 and 1000 and it's going to have a strong effect on the amount of stylization applied to your image. Okay, so we will start with this simple prompt. Medium, photo, steampunk, male portrait looking at the camera close up and. And we will set the default value to 100. Okay, so these are the results that you get when we set stylize to 100. And these looks pretty amazing images. Right now I change it to zero. The images are pretty bland, not great at all. But actually when I set it to 1000, you see a lot more details and you can see that there is a lot more creativity in the generated images. This was a quick roundup of all the features that are added in version 5.2. I'm especially interested in the zoom out and the shorten command. Now it seems like Midjourney is going to be adding a lot more interesting features. Unfortunately, Midjourney is not free anymore. If you're looking for some alternatives, check out Leonardo AI and Playground AI. They offer some really good image generation tools for free or close to free. I hope you found this video useful. Consider liking the video and subscribing to the channel. Subscribe if you are interested in Generative AI both for image generation as well as for large language models. Check out the description of the video for link to the Discord server. You will find like minded people over there who are helping each other out, so come join us. Thanks for watching and see you in the next one.