New Anime Model Added to System
Hello there, so Midjourney just announced one of the big additions to the system, which is the ability to render anime style. If we look at the announcement, we can see that they have released a brand new anime model which can be accessed by using the suffix “niji”, which means rainbow or 2D in Japanese. This is supposed to work well for an anime or cartoonish style. So, what I want to do is go test this on Discord and see how it works compared to a local installation of stable diffusion with our anime model for stable diffusion. We’ll go and compare how both of them work together and see how they compare or what is different.
Testing the Anime Model on Midjourney and Stable Diffusion
So, some tests have already been tried and when I saw this I had to play with it. Okay, and I won’t go show you something good that I started and something we’re going to do together during the video. Here’s the first thing I tried to do. It says “beautiful girl in Victorian dress laid down on a field with wind move your hair leaves flowing around top down view”. And this is just the normal version 4 without any other stuff and you can see it does have some illustration effect. It’s not totally photorealistic, but for this I need the AdWords photorealistic. But it is a nice, beautiful look.
So the next thing I did was exactly the same, only I added the anime style by adding the “niji” tag. So we’ll just go to anime style and we can see it does apply somewhat of an anime style. It’s a nice looking image and it can work very easily as an anime style. And right here is with a ninja attack. You can definitely see the difference. It’s a little bit more stylized and a little less sterilized, but it does look a little bit different. That’s a mix with this.
So let’s see what happened with stable diffusion. Right here we have stable diffusion version 3.0, which allows us to create anime using the same string. I used it with 20 steps, but it can go up to 50 probably. It’s a little bit better, but the point is right here you can see what it produced for us. It does have a little bit of a Victorian Steampunk coloring, very washed out black color. But right here you can see we have this image, it’s not as good as what I produced, but it isn’t very much an anime style.
So let’s go to the next one, which is a bunny knight fighting fire breathing dragons. And of course this is our anime style. And what happened is the bunny is just roasting another bunny. Okay, so this is if you just type “anime style” right there, which is actually not necessarily anime style. But this is the reason for this string I was using because it’s hard to reference. If we type in “definitely” if we just go and type “imagine” okay, um, “Dragon Ball Z Super Saiyan”, okay “Super Saiyan”. And it does help when you can actually spell properly. So I spelled it wrong, come on, right here saying “Dragon Ball Z”. Let’s go add it like right there, yeah capital there you go, let’s copy this one.
Okay, actually before we press let me go see what’s done here. And this is actually what unstable diffusion gave us. That is right here, I can see this. Not the ears, rocky ears, yeah. That is a bunny and a dragon. So actually the stable diffusion model performed better with this mixing than the Discord “mid Journey” because this is what the mid Journey gave us on the same strings. Versus what
is a stable diffusion with the anime given to us. A bunny fighting a dragon with fire breathing. So you can see all of this here, actually they performed better than this okay.
So let’s go back now to our “Dragon Ball Z Super Saiyan”. It seems like it’s good, let me go copy this. Okay, and we’re going to do the first one normal. Okay, let’s go next we’ll do this with a photorealistic tag. Okay, we’ll do the same one “Dragon Ball Z Super Saiyan”. And we’ll do the same thing with the anime style. So here’s the normal one, here’s the photorealistic, and here’s the anime style. And you can see the difference, the normal one is a little bit more realistic, the photorealistic is even more realistic, and the anime style is a little bit more stylized.
Conclusion on Performance of Anime Model on Different Systems
So, overall it seems like the stable diffusion model is performing better with the anime style than the Discord “mid Journey”. It’s able to produce more of a true anime style image than the “mid Journey” was able to. But, it’s important to note that both of them are able to produce a nice looking image and it just depends on what style you prefer. So, that’s it for this test. Thanks for watching and we’ll see you in the next one.