Hello, I'm Choi Donhyeon of
Stable Diffusion Korea and soy.lab. Nice to meet you. Today, let's take a look at the new
update function of Midjourney V6, SREF. This time, we're going to talk
about this content so easily. Before I explain this, I'm going to tell you a good news. This time, I and Midjourney bestseller Cho Namkyoung will launch a Midjourney professional
lecture on Fastcampus X soy.lab. Wow, I will give you the document and QR code together. You can scan the QR code here or click the QR code in the document to access it. We will officially share related links when this video is released. We will also be distributing discount coupons,
so please show your interest. You don't have to pay attention. The discount coupon will come out soon, so click on it right away and enter the payment button right away. Now, let's take a look at today's
content, Midjourney V6's style reference. It was February 1, 2024. In Korean time, the content related to
the morning of 8.45 a.m. was updated. These functions have been updated. To be exact, this consistent style and this style reference are going to be updated soon. This time, the first updated function is the content related to this style reference. I'm using this sREF function. And if you add an additional reference to the image path, you can use the related functions. The contents are included here. The important thing is
that the related content can be used in Midjourney V6 and Niji V6. Currently, parts related to this feature do not work in V5. And now, if you look at the advanced settings related parts, the contents that you can mix up to
three maximum images are included. The fact that you can give each
related value separately and that you can basically raise the style weight value
from 100 and up to 1000 are included. So today, I'm going to
proceed with this content. I'll try to proceed based on the content. Let's start Take 1. I created an image in
Midjourney like this. With this image created, I'm going to see the
process of creating an image using the prompt and this
sREF function as a result. First of all, you can proceed with the settings related to this prompting in this way. Execute the creation
command called imagine. We will set the relevant prompts, sREF function, path, weight default and stylize function to 1000 and proceed in Niji. But if you look at it like this,
it's a little hard to recognize. I will explain it by dividing it into blocks to make it easier to read. So, activate the function called imagine. Activate this basic prompt here. Apply the sREF related settings here. Add additional setting parameters. Now, let's go ahead and set it up. First of all, to get
this related image link, Please upload the
image I created like this. After uploading like this, all you have to do is get the link to the related image. I just copied the image address like this. If you paste the address, you can
see that this kind of address is created. It's a little long. But this address will be reduced later. If you look at the prompt settings we provided in the document, There's an address in here. You can change this
address to this address. It's been changed. Now activate the imagine prompt and create an image by copying and pasting the prompt below. Then, in this way, You can see that it's starting right now. In the case of MidJourney, When you create it, You won't see the seed value here. Of course, you can see it separately. The important thing
is that when you create it every time, Because
random is the default, If I create this like this, It's going to be a little different from
the image I showed you as a reference. The result will come out. You can see that the image
has been created like this. It's coming out well. If you look at it now,
In this anime style, You can see that the
image here has been created. You can see what I created before. I created it like this before. You can see that the prompt here and
the prompt here are completely the same. There were four images created at the time. And I chose one of them. The feature is that the
prompt here is very simple. To describe the actual image here, I should have added more
information about prompting. Currently, based on the
image I just referenced, You can see that the
content is well composed. The image that was
made like that is this image. What I wanted to express was the diffusion of the current trendiness and the feeling of the mid-1990s. You can see that the
feeling is expressed very well. Let me explain this a little more. Basically, this is the current state. You can see that the image applied as the prompt on the left
and the style reference on the right has been composited into the image in the middle. If you compare this with the prompt
composition we checked just before, Here you can see the 1980
anime style delivered this way. Although it is an old anime style overall, you can see that a trendy feel is reflected. And through the sREF value here, you can see that details such as those in the tree's shadow, pose, gaze direction, and range of exposure to light are similarly applied. So, even if the prompt is simple, we can see that we can easily create the picture we want to create. So, if you use these
features a little more, You can try to create a
variety of pictures like this. If we were just trying
to describe these parts, some keywords, such as window lighting, must be included in the prompt to be displayed. Or, keywords such as tree shadow should be included. Now, through the sREF function, Reflect the scenarios in it. Or you can see that you
can apply parts related to style. Now, the second one is... Among the things we
talked about at the beginning, I explained that you
can mix various images. I'm going to mix and use
parts related to this style. I'll show you. This time, it's a little bit more... I've created an image with an old color. This image was also created at Midjourney. And with the image I made at the end, I'm going to do a style mix related to it. We plan to fix the prompt we used earlier and
mix the three pieces of data for diffusion. The prompt used here is like this. I'll make this a little simpler. If you look at it, it looks like this. You're in an order with the
existing prompts and sREFs. The link is related to number one. And a new link is going in. Number two goes in. So just remember this related part. If you want to add one more image, Then you can get up to three as much as possible, and if you put these related parts in,
the style mix will proceed. So if you look at the results made, you can see these results. So if you compare it a little easier to understand, you can see that the 1st image made in real life and the 2nd image of a slightly older style are made here. If you mix the related parts
through the style mix in SREF, the parts related to 1 to 1
are currently being discussed. The parts related to the hairstyle and color here, the
parts related to the lighting scenario here, and the parts related to the mood are mixed moderately, so you
can see that you are trying to maintain this middle value. If you look at it now, you can see that the
lighting blocker is well inserted like this, and you can see that the feeling of the curly perm
related to the hair is expressed in this way. You can see that it is
expressed in this way. Let's take a look at how
this part was actually applied. First of all, let's call the
result value that came out. At the time, the result came out like this, and if
you zoom in on what you can see here, you can see that the 1st and 2nd images have been referenced
through the SREF function of this basic prompting. In fact, it's a bit meaningless,
but SW100 is included as the basis, and the stylized value 1000 is
currently included in this way. Image created with Niji 6. But what's important is that this image we're looking at right now is a 1 vs 1 mix. What I'm talking about is
that the situation now means that the images 1 and 2 are
just mixed with a 1 to 1 ratio. If you check the first notice box
right now, the proportion of this style reference related part can
be adjusted to the desired shape. It was written. So, you can adjust the style mix by adjusting
the ratio very strongly in this way. So, in order to apply this related
weight value to each image, you can adjust the weight value by using
two columns in this way and adding numbers. In fact, related parts are also used to adjust the weight of each character when prompting in Midjourney. So, you can adjust the ratio
of each image by adding two columns and adding a number
at the end of the column. So, you can adjust the ratio
of each image by adding two columns and adding a number
at the end of the column. So, if you look at this part,
you just need to raise the weight value related to the
image you want in this way. So, if you assume that you applied it
this way, you can see the result like this. Let's call the job about
this content and check it out. So, you can see the result
value when applying four Weight 4. You can also check the
result value of applying the weight value of the old color
style image from the back. So, to summarize this one more time, the writing scenario or the
story related to the composition of the actual image that
we made at the beginning. Actually, this is the whole
style, but that's what it's about. And another style. It's a completely
different style right now. Create an old-fashioned anime
cartoon image that's in there. If you adjust the sREF weight
value in the form we want, You'll be able to see this kind of result. So if you increase the weight
value here a little bit more, You can see that the
parts related to the perm hair and curling hair
will definitely appear. And on the other hand, if the weight
value related to the real thing is high, this blocker in a lighting scenario. You can see that it's
applied to all images. Instead, the parts related to
this hairstyle are a little bit... You can see that it's changed a little bit. In the Midjourney, there are really a lot
of ways to fix or maintain these parts. Anyway, if you use this sREF function well, the continuity related
to the style is through this function. By referring to it, we can
make the pictures we want. I was able to check that. Here, if the cREF function is added later, It's going to develop into
a really amazing function. Yes, that's all I've got for you today. We've learned about the
sREF function of Midjourney. In the future, we'll be able to deal with
the parts related to Midjourney together. Don't forget to subscribe and like. I'll be back with good content next time. Thank you.