If you’re here, you have no doubt seen examples of some of the mind blowing capabilities of Midjourney to render characters in pretty much any style you can imagine. You may have even created your own already. If you’ve had a chance to take Midjourney for a test drive, you will have quickly realized that, even with the best prompting skills, it produces randomly inconsistent characters by design.
If you’re looking for a one-off image, you can easily produce stunning results. But if you are looking to create, or use a character you’ve created, that’s “reusable” in different situations, you’ve probably figured out that Midjourney was never developed to create consistent characters.
You can try the same prompt a hundred times and it will give you 100 different returns with 4 different images each time, none of which look like the other. This becomes an issue if you want to create a character for something like a comic book, children’s story, video game, graphic novel or advertising campaign in which you want a character to be consistent throughout.
In this article, we will show you how to create consistent characters in Midjourney. It’s not the easiest task, and if you’re not wedded to Midjourney, check out our guide to creating consistent characters in Scenario.
We’re going to split them up into 3 groups – photorealistic images/realistic illustration, anime/comic/storybook and 3-D Pixar style characters. You’ll find that different techniques work better with different styles, and we’re going to show you what works best with all three.
Creating Consistent Realistic Characters in Midjourney
Assuming you have setup an account on Discord and signed up for Midjourney, it’s time to login.
We’re going to go over two ways to create consistent characters in Midjourney: the “Seed” method and the “Vary Region” method. The “Vary Region” method is easier, but it’s helpful to learn the Seed method first to understand how everything works before moving to the Vary Region method.
Let’s start with the Seed method.
Consistent Photorealistic Characters in Midjourney: Seed Method
Once you are in Midjourney, the first thing you want to do is open up a word or notepad document. We’re going to be using that to cut and paste prompts and image links we are going to be using throughout the process.
Step #1: Create your character and give it a name.
Make the name a part of every prompt, so you have one more element that associates your character with the additional renderings of the same character you’re going to do.
Unlike when you are creating a one-off character for a specific purpose, you should keep your prompts short. The purpose of the initial prompt is to get a character looking like you want it, without worrying about what it’s doing or how it’s feeling, or any sort of setting/background. Your initial character is what you are going to be basing your future renderings on of the same character.
First, start with the name you’ve given your character.
Then add the description of what you want your character to look like. You can add a description of what they’re wearing if it’s important to the character’s identity, like for example, if they are wearing glasses, a uniform or will wear something that will be a constant throughout that will become their ‘trademark’.
Then you can add other prompts to specify things like the version of Midjourney you want to use, as well as qualitative, lighting or photographic elements to your prompt. See this article for a detailed guide on How To Create Photorealistic Images With Midjourney.
Side Note: Choosing your version of Midjourney will impact your results. If you are going for ultra-photorealistic characters, you may want to give version 5.x (–v 5) a go to see how the different versions compare with each other. Version 4 will give you a realistic illustrative style, while version 5 will likely produce what looks more like an actual photograph of a real person. The version that’s right for your character really depends on what you’re looking for.
Your initial prompt should look something like this:
Prompt:
“Erin Byrne, photo of a beautiful fierce warrior woman with braided red hair, green eyes, and freckles, standing on a platform in a castle, cinematic lighting –ar 2:3“
Once you’ve written your prompt, hit “enter” and wait for Midjourney to render your images. You may not get the character you wanted the first time around. If it was not even close, you can always adjust your prompts and try again.
If there is an image that is kind of leaning in the right direction, you can opt to choose that image and get Midjourney render four more variations loosely based on that image using the V1 – V4 buttons, and it should result in images that are closer to what you’re looking for.
In this case, I liked the third image, so I created variations based on image 3.
In the event that there’s an image that’s almost exactly what you were looking for, but you just need to make a few tweaks to your prompt, you can choose the ‘remix’ setting. This will give you 4 more variations of a particular image, and will also allow you to edit the prompt.
Step #2: Upscale Image And Get the Seed Number
Once you get an image of your character from Midjourney that you are satisfied with, upscale it by using the U1, U2, U3 or U4 button that corresponds with your chosen image. In this case, I chose image 2.
When you get the upscaled version of your image, click on it and select the “open in browser” link.
Right click and copy the image address. Now remember at the beginning we said – first open a word or notepad document? – this is where you’re going to paste that image link, because you’re going to need it later.
Now you are going to find the seed number of your image. The seed number is a reference number to the image that Midjourney will use to create other images based on that original image, so this is an important step.
To get the seed number of the image, find the upscaled image on the Midjourney feed. Click on “add reaction” which is in the little smiley face emoticon with the “+” on it. You’ll find this icon if you scroll up and hover your mouse at the top edge of your image.
This will open up a box. Click on the envelope (you might need to search for the word “envelope”). What that does is generate the same image on the feed, only this time it shows the seed number.
Copy and paste your seed number where you pasted the image link.
Step #3: Create a second image consistent with your first character
To create an image of the same character doing something different, first paste in the image link into the prompt box that you got from the image you upscaled.
Then, paste in the exact same prompt you used to create the initial character. Do not change the wording of the core prompt.
You can add in whatever extra actions or changes in clothing, scenery, or positioning, then add the seed number(–seed 1234), where ‘1234’ is the seed number you’d copied from your original character image, at the end of the prompt.
Your new prompt should look something like this:
[image link] Erin Byrne, photo of a beautiful fierce warrior woman holding a sword, with braided red hair, green eyes, and freckles, standing on a platform, cinematic lighting –ar 2:3 –seed [1234]
The text that’s bolded above is your core prompt. For every subsequent image you make in Midjourney of the same character, you will use:
- the original image link (at the beginning, followed by…)
- the core prompt (unchanged, and…)
- the seed number of the original image at the end of your prompt.
You can change all of the other stuff in between to whatever you want your character to be doing or to adjust lighting or scenery or clothing, etc.
Here’s yet another set of images in the series, with the prompt:
[image link] Erin Byrne, photo of a beautiful fierce warrior woman sitting by a campfire in a long dress, with braided red hair, green eyes, and freckles, cinematic lighting –ar 2:3 –seed [1234]
It’s a good idea to copy/paste that prompt where you put your image link and seed number and save it because you are going to be using that same prompt every time you want to create a new image of your character.
I also just want to note that not every rendering will look exactly like your character, but you should get at least one or two of the four that look pretty similar.
Every time you get another good likeness of your character, save the image link, like you did with the first one. And the next time you create the character, add another image link.
This way Midjourney has a bunch of image references of your character in different positions and poses, doing different things that look similar.
This all helps to get even more consistency each time you generate a new image of your character.
Consistent Photorealistic Characters in Midjourney: Vary Region Method
Midjourney added a new feature in August 2023 called “Vary Region” that is a real game-changer. It allows you to vary just a section of an image, while keeping the rest of the image intact. You can edit your prompt or use another image as a reference image for the region.
This makes creating consistent characters much easier than the Seed method described above.
I wanted to create an image of Erin Byrne from the Seed guide above fighting with an Orc creature. I created the image below:
Cool image, but clearly not the same Erin Byrne from the Seed section.
To remedy that, I upscaled the image, and clicked on the Vary Region option. This opens a new window in discord, which allows you to select the area of the image to edit.
You can use a box selection or the lasso tool. I used the lasso tool to select her face and hair, like so:
I included a link to one of the previous Erin Byrne images used above, and left the rest of the prompt the same. The resulting image is here:
Not bad, and much easier than the whole Seed thing! Using multiple images or specifying poses can help, but this method is quite powerful on its own.
Creating Consistent Anime, Story Book & Comic Characters in Midjourney
You can actually try the method above for more illustrative styles of characters, but there’s a much easier and, some would argue, more accurate way to create a character and have it consistent from different angles and in different poses, which unfortunately does not work as well in Midjourney, for the more realistic characters.
When you are working with an illustrative style of character, like comic book, children’s story or anime characters that are a flatter, 2-dimensional style, you can create a character sheet.
Step #1: Create a Character Sheet
Creating a character sheet is a great way to show how your character looks from different angles in the same image. Don’t forget to give your character a name!
Prompt:
Kiko Sushi, character design sheet, cute, young japanese girl in a red traditional Kimono, japanese woodcut, flat color, full body, multiple poses and expressions, character concept, –no outline –upbeta
Side Notes:
- The Beta Upscaler is ideal for simple, flat illustrations. When you use the –upbeta parameter, instead of the U1 U2 U3 U4 upscale buttons upscaling the image to the default of 1024px X 1024px, your image becomes 2048px x 2048px.
- You can choose which version of Midjourney to use. v5.x is typically better for realistic images, but v4 tends to be better for comic book, cartoon and children’s book characters. But if you are creating an anime character, try –niji at the end of your prompt. The niji model is specifically designed for anime style art renderings and might garner you even better results.
- Copy/paste your prompt into a word or notepad document because you’re going to use it later.
Midjourney will render you 4 images with 5-6 characters per image. You’re going to upscale them, cut them out and separate each individual image. In my case, I used the top left set of images.
Then you’re going to use the individual images as references to get even more consistent images of your character.
Step #2: Upscale, Cut Out Your Characters & Paste Them Back Into Midjourney
Click on the upscale button that corresponds with the image you like. Click on the upscaled version of the image.
You are going to require an image editing program, like Photoshop for this next part…
Right click and copy the image of your character sheet and paste it into an image editing program. Here, you’re going to want to get rid of any unrelated extraneous things, like random stuff in the background or around your character that doesn’t belong.
You may also find that not all of the renderings are fantastic. Get rid of the ones that you don’t like.
For the remaining characters in your character sheet, you are going to select them individually, one at a time. Once you have one selected, just copy the selection and paste it into the Midjourney dialogue box at the bottom.
Do not type ‘/imagine’ to create a prompt box when you are pasting them back in. Just paste the image directly into the text message box, or click the + icon and add your images.
You can insert them all in one shot, or upload individual images per line.
Step #3: Save The Image URLs to Notepad
Now you have all of the different variations of your character uploaded back into Midjourney, you’re going to click on the first image and then right click to select ‘copy link’.
Paste the image url into your word or notepad document. Do this for each of the images you’ve uploaded.
Step #4: Create A New Prompt For A New Character Action
In Midjourney, type in “/imagine” and hit the spacebar to pull up the prompt window. Paste all of the image links into the prompt window with a space between each link. Then copy and paste your initial prompt. From now on, your future prompts will be consistently tied to the aesthetic of this character!
Remove any mention of “character (reference/design/concept) sheet” from the original prompts. And where you have “multiple poses and expressions,” replace that with another action you want your character to do, like running or dancing or waving. Your new prompt should look something like this:
“[image link 1] [image link 2] [image link 3] [image link 4] [image link 5] [image link 6] Kiko Sushi, cute, young japanese girl in a red traditional Kimono, bowing, japanese woodcut, flat color, full body –no outline –upbeta”
Where it says [image link], you just need to paste in the image links without the brackets and leave a space between each of the links.
Hit ‘enter’ on your keyboard and Midjourney will render 4 more images of your character doing what you specified. I used bowing in my example.
Repeat this process for whatever action you want to see your character perform.
In the example above, it seems like my Kiko bowing is a little older than the original. You may need to run a couple of iterations to find the right image.
Here’s another version, with Kiko waving, using the same process.
Creating 3-D Pixar Style Characters in Midjourney
Okay so the good news about 3-D Pixar style characters is that you can use either of the methods above to generate consistent characters.
While the –seed method seems to work best for photorealistic images and realistic illustrations, and character sheets are more suited to 2-D, flat illustrations, both methods seem to work equally well with 3-D Pixar style characters!
All you have to do is create a 3-D Pixar style character prompt, like: “Tezza, character design sheet, an electric pangolin, Pixar style illustration, multiple poses and expressions, character concept, white background, –no outline –upbeta –v 4” and follow the instructions above.
Pick which set you like the best, cut them out individually, and paste them back into the discord chat. From there, you can iterate more expressions and actions.
Take Away
If you are looking to create consistent characters, the “–seed method” works best for photorealistic and realistic illustrated characters, and character sheets work best for 2-D, flat illustrated characters. Both methods can work with 3-D Pixar style characters.
Midjourney was never meant to create consistent characters, which is why it requires so many steps. But with a little know-how and some prompting finesse, you can manage to not only create a character in any style, but also make multiple generations of the same character doing different things in different situations that are relatively consistent.
