How to Use Multiple Characters in Midjourney

Are you struggling to place multiple consistent characters in a scene with Midjourney?

Well, you’re not alone and it remains one of the biggest unsolved mysteries within the Midjourney fan community.

But after hundreds of requests, I decided it was time for me to give it shot.

This is a story of many failed attempts but also a silver lining on the horizon.

In this guide, we’ll show you which methods definitely DO NOT work but also provide insights into what DOES!

Let’s dive right in.

Like this content? Then share it!

How to Use Multiple Characters in Midjourney
Here’s the answer. Great article by @ChrisHeidorn right here: https://tokenizedhq.com/multiple-characters-in-midjourney/

www.tokenizedhq.com

How to Create Multiple Characters in Midjourney

This guide is part of a larger series that covers character design in Midjourney.

If you haven’t read Parts 1, 2, and 3 of our series, I strongly recommend you check them out.

They cover everything you need to know about creating your very first consistent character in Midjourney, placing your character in action scenes, and applying a consistent style to your character.

Reading through all of the previous parts will provide valuable context on what I’m about to show you in this part of the series.

Before we jump in, I must confess that I wasn’t sure whether it was possible to place multiple characters in Midjourney in a consistent way.

Placing multiple characters in an image is one thing but making sure they remain consistent is a whole nother story.

Naturally, I had to find out whether it could be done.

And just so you know….I’m not promising you anything.

But if you’re curious to find out what fascinating things I learned, then come along for the entire ride and keep on reading or check out the video on YouTube.

You will not regret it.

Complex Characters with Constructed References

So for this exercise, I’m going to use our good old friends Carla Caruso and Peggy Palermo.

I will try to place Carla and Peggy in the middle of a New York City street and I also want them to raise their hands with a fist as if they were protesting.

I’m going to start off by recovering one of the original image references we created in Part 3 of this series.

/imagine glamour shot full body image, beautiful woman, bright red braided bun hair, only red hair, wearing a leather jacket in gray color, wearing a white tank top under jacket, wearing beige pants on legs, in the style of a Marvel comic book --ar 2:3 --seed 2385530314

This was one of the character concepts that we used to blend into a new one.

Then I’m going to create a similar type of full-body character concept for Peggy.

/imagine glamour shot full body image on white background, beautiful woman, bob cut, pink hair, blue eyes, black leather clothing,
black pants on legs and black boots, in the style of a Marvel comic book --ar 2:3 --seed 2936472758

Now that we have our source character concepts, let’s create new images of Carla and Peggy in the protesting pose that I want.

I’ll start with Carla.

/imagine https://s.mj.run/rpab9si_RhA beautiful woman, bright red hair, braided bun hairstyle, only red hair, marvel comic book style, holding a fist in the air, raising her arm in a fist --ar 2:3 --seed 3832683861

And then we’ll do the same for Peggy.

/imagine https://s.mj.run/YeBOPlpykXQ beautiful woman, bob cut, pink hair, black leather clothing, in the style of a Marvel comic book, holding a fist in the air, raising her arm in a fist --ar 2:3 --seed 4069949583

Ok great.

Now that we have our character references, let’s create some background scenery where we can place our multiple characters in Midjourney.

/imagine panoramic view of a long street in new york city, dystopian theme, Marvel comic book style --ar 3:2 --seed 789374466

Excellent, these are just what I need.

I’m now going to create 3 different image references inside my image editor.

In each one, I’ll use a different background image as well as different versions of the characters.

In order to do this, I had to remove all of the white backgrounds of the character references and then place them on top of the background images.

I’m hoping that the image references will be dominant enough for the AI to understand what I need and create an image that shows multiple characters in Midjourney.

So let’s blend all 3 of them together.

/imagine https://s.mj.run/bHe7lSlculQ https://s.mj.run/lNjO6oTNhwU https://s.mj.run/0F9ya4u4HEU

Yikes, those turned out to be absolutely hideous.

So maybe I do need to add a text prompt after all?

Let me add some more context and see if that works.

/imagine https://s.mj.run/bHe7lSlculQ https://s.mj.run/lNjO6oTNhwU https://s.mj.run/0F9ya4u4HEU beautiful woman with bright red hair standing next to another attractive woman with a pink bob cut hairstyle, red-haired woman is wearing a white tank top and gray leather jacket, pink-haired woman is wearing black leather clothing, marvel comic book style, holding a fist in the air, raising their arm in a fist --ar 3:2 --seed 1426174332

Ok, so if you look at the images it creates, it’s quite clear that this is not working.

So what’s wrong here?

• It seems that the image references are able to convey the correct composition.
• But Midjourney can’t assign the correct hair to each character.
• It also can’t really assign the correct outfit.
• And the faces are just plain ugly.

So this attempt was a complete FAILURE.

Maybe the characters are just too complex and placing multiple characters in Midjourney requires simpler character concepts.

I recently saw another YouTuber presenting his approach to creating consistent characters.

And he kept using very simple, cute, and small characters.

Maybe it just works better with those?

However, he was just creating character concepts and didn’t actually try to place multiple characters in a scene.

Either way, let’s give try that.

Simple Characters with Constructed References

So let’s start off by creating 2 different character concepts.

First, I’m going to create a cute little mouse with an ascot cap.

/imagine character concept on white background, multiple poses and expressions, little white mouse character wearing a brown ascot cap, children's book illustration --ar 3:2 --seed 3428420958

I really like the bottom-left set, so let’s go with that one.

My second character is going to be a cute little bear wearing a red shirt.

/imagine character concept on white background, multiple poses and expressions, little brown bear character wearing a red shirt, children's book illustration --seed 3748491214

The bottom right one is just perfect for this use case, so let’s keep that set.

Next, we need some background scenery for our characters.

So let’s create an image of a children’s room.

/imagine ground level shot of a children's room, children's book illustration style --ar 3:2 --seed 3669121037

Great, let’s pick the bottom right one.

I then bring everything back into Photoshop, cut out my characters, and place them on top of the background.

If it wasn’t for the different drawing styles, you could almost use this as an actual illustration for a book.

You might be tempted to just blend together the background image and this new composition reference we created.

And this will reproduce your scene in a consistent style, but once again, it will mess up your characters.

Anyway, so let’s try blending our reference image with a prompt.

/imagine https://s.mj.run/8EbMySNNlc4 ground level shot of a children's room, little brown bear character wearing a red shirt, little white mouse wearing a brown ascot cap, a bear and a mouse having a conversation, children's book illustration style --ar 3:2 --seed 1455446854

This creates a set of images that mostly represent the concept that we had in mind and the characters do look good.

But they’re simply not the characters we wanted and their features are all over the place.

And remember, the objective is not to simply place multiple characters in Midjourney, we also want them to look consistent.

So what does this tell us?

• Simpler characters are a bit easier to create
• Placing two or more in the same scene is still very difficult
• Midjourney isn’t able to assign the correct features to specific characters
• You mostly get mashups of your ideas

Ok, so as much as it pains me to say this but I think image references are not going to get us very far.

As soon as you want to use more than 1 specific character in a scene, the lines start to blur and concepts invariably bleed into each other more than you want them to.

So this attempt was also a complete FAILURE.

Let’s move on to multiprompts and gives those a try.

2 Characters with Multiprompts & References

Multiprompts have the major advantage of allowing us to reinforce specific concepts from different points of view.

So that’s why I’m hopeful that this might work.

However, I also know that they have a similar “bleeding” problem when it comes to characters that belong to the same category.

This is particularly obvious when you try to place 2 different animals next to each other.

Have a look at this:

/imagine a brown bear sitting next to a sloth --ar 3:2 --seed 1351973951

I think it’s quite obvious what the problem is.

Midjourney keeps creating weird fantasy creatures that are a blend of both animals.

I also have no clue what these bird-like creatures are doing in the image.

Things are considerably easier when you try to place two completely unrelated character types into an image.

And by unrelated I mean animals and humans.

Let’s start off with something simple.

/imagine cute brown bear speaking to a little boy --ar 3:2 --seed 811417316

This concept is crystal clear to Midjourney.

Now let’s add some color to the boy’s clothes.

/imagine cute brown bear speaking to a little boy, little boy wearing blue clothes --ar 3:2 --seed 35538780

So far this is only a single prompt, so let’s add a second segment in order to reinforce the concept.

/imagine cute brown bear speaking to a little boy, little boy wearing blue clothes:: a conversation between a little boy in a blue outfit and a cute brown bear:: --ar 3:2 --seed 2645177336

As you can see, anytime you add color into the mix, there’s is always going to be some color bleed.

Sometimes you can fix this by adding a negative prompt like this.

/imagine cute brown bear speaking to a little boy, little boy wearing blue clothes:: a conversation between a little boy in a blue outfit and a cute brown bear:: blue background::-0.5 --ar 3:2 --seed 3013711331

While this isn’t perfect and it sometimes removes more color than you want to, it’s still quite useful to know.

So let’s change the location of our characters and place them on a farm.

/imagine cute brown bear speaking to a little boy, little boy wearing blue clothes, country farm in background:: a conversation between a little boy in a blue outfit and a cute brown bear, background scenery of a farm:: --ar 3:2 --seed 140096663

So do you notice an obvious problem here?

Picking a farm was probably the dumbest thing I could do.

What do you usually find on a farm?

That’s right, tons of different animals.

And that’s why our bear looks like he’s transformed into a pig-like or cow-like creature in the two bottom images.

Once again, Midjourney fails at keeping characters of the same category apart.

So clearly, we need to move to another location.

/imagine cute brown bear speaking to a little boy, little boy wearing blue clothes, garden in background:: a conversation between a little boy in a blue outfit and a cute brown bear, background scenery of a garden:: --ar 3:2 --seed 289742806

Alright, so this is much better already.

Now let’s inject some children’s book illustration style into the prompt by adding it only to the first segment.

/imagine cute brown bear speaking to a little boy, little boy wearing blue clothes, garden in background, children's book illustration:: a conversation between a little boy in a blue outfit and a cute brown brear, background scenery of a garden:: --ar 3:2 --seed 3684279999

If you want to get an even stronger children’s book style, you would simply add the same style to the second segment as well.

/imagine cute brown bear speaking to a little boy, little boy wearing blue clothes, garden in background, children's book illustration:: a conversation between a little boy in a blue outfit and a cute brown bear, background scenery of a garden, children's book illustration::
--ar 3:2 --seed 2049437021

Sometimes the increased abstraction of a children’s book illustration can be helpful, so I’m going to try something on a whim.

I like the top-left and bottom-left styles, and I’m going to try to apply them to a slightly different scene.

Let’s start off by creating a new prompt without image references.

/imagine cute brown bear and a little boy sitting on a bench, little boy wearing blue clothes, garden in background, children's book illustration:: little boy in a blue outfit and a cute brown bear sitting on a bench, background scenery of a garden, children's book illustration:: --ar 3:2 --seed 3658801722

Now, you might really like the style of these images, but try to ignore that for a moment.

The only thing we’re going to do right now is to confirm that our text prompt actually gives us the type of composition that we want.

And clearly, it does.

So now let’s let use that very same prompt and add one of the image references as an image prompt.

/imagine https://s.mj.run/dnDW1rkVd9w cute brown bear and a little boy sitting on a bench, little boy wearing blue clothes, garden in background, children's book illustration:: little boy in a blue outfit and a cute brown bear sitting on a bench, background scenery of a garden, children's book illustration:: --ar 3:2 --seed 3310766378

Now, this is something we can potentially work with.

Let’s try this with the other image reference.

/imagine https://s.mj.run/49MvKMDgfRM cute brown bear and a little boy sitting on a bench, little boy wearing blue clothes, garden in background, children's book illustration:: little boy in a blue outfit and a cute brown bear sitting on a bench, background scenery of a garden, children's book illustration:: blue objects::-0.5 --ar 3:2 --seed 470195360

I personally think these are awesome.

So I think it’s fair to say that this one is at least a partial SUCCESS.

It’s crucial that you do not get too hung up on minute details.

Ignore the micro, and focus on the macro.

If you plan to use this in a children’s book, let’s be honest here.

The majority of people out there are not going to notice tiny differences, as long as you get the basics right.

So what have we learned here?

• Avoid placing more than 1 animal in your scene
• An animal and a human are fine
• Avoid color bleed by fine-tuning with negative prompts
• Focus on getting the initial text prompt right
• Then apply styles with image references
• Tweak your text prompt to change the composition and dynamic

So, what about 3 characters? Is that something we can do?

Well, let’s try it.

3 Characters with Multiprompts & References

I’m going to skip the simple prompts and jump straight into a more complex multiprompt with 3 segments.

/imagine image of a garden, a little boy and girl with long blonde hair having a conversation with a white rabbit, garden in the background:: garden scenery, girl and boy speaking to a little white rabbit:: a little white rabbit speaking to two children in a park:: --ar 3:2 --seed 3701118171

So these actually look pretty good.

We have an animal, a female character, and a male character.

I’m going to avoid adding colors to the prompt, simply because I know that it will result in color bleed.

I also won’t add any more detail to any of the characters, simply because I know that the concepts will lead to strange results.

For example, if you introduce the concept of a “dress”, then the boy is more likely to have feminine traits or turn into a second girl altogether.

Instead, I’m just going to add “children’s book illustration” to all 3 segments of the multiprompt.

/imagine image of a garden, a little boy and girl with long blonde hair having a conversation with a white rabbit, garden in the background, children's book illustration:: garden scenery, girl and boy speaking to a little white rabbit, children's book illustration:: a little white rabbit speaking to two children in a park, children's book illustration:: --ar 3:2 --seed 3997520679

These look really nice.

Now let me try to apply the same logic as in our previous attempt.

/imagine image of a garden with a pool, a little boy and girl with long blonde hair sitting next to a pool with a white rabbit, garden with pool in the background, children's book illustration:: garden scenery with a pool, girl and boy speaking to a little white rabbit sitting next to a pool, children's book illustration:: a little white rabbit speaking to two children in a park with a pool, children's book illustration:: --ar 3:2 --seed 1259112591

Ok, so the concept isn’t perfect but it’s close enough for my taste.

Let’s see what happens when we use this very same prompt with the image reference that we picked earlier.

/imagine https://s.mj.run/OwIcRvLGUb4 image of a garden a pool, a little boy and girl with long blonde hair sitting next to a pool with a white rabbit, garden with pool in the background, children's book illustration:: garden scenery with a pool, girl and boy speaking to a little white rabbit sitting next to a pool, children's book llustration:: a little white rabbit speaking to two children in a park with a pool, children's book illustration:: --ar 3:2 --seed 2453499868

BOOM!!!

Now that I believe qualifies as a SUCCESS!

But I actually wanted to go one step further.

Manually Adjust the Color of Clothes

I wanted to see whether I could keep the style and composition, but change the color of the characters’ clothes.

So I took the original reference as well as one of the new images and added some strong color to their clothes.

I then use both images as a reference and simply repeated the rest of the prompt.

/imagine https://s.mj.run/tzkt9dsWKRQ https://s.mj.run/1hMXkzpCuMA image of a garden a pool, a little boy and girl with long blonde hair sitting next to a pool with a white rabbit, garden with pool in the background, children's book illustration:: garden scenery with a pool, girl and boy speaking to a little white rabbit sitting next to a pool, children's book illustration:: a little white rabbit speaking to two children in a park with a pool, children's book illustration:: --ar 3:2 --seed 3131313437

And here’s what I got.

The boy’s clothes still aren’t very consistent and they keep changing in every image.

However, I successfully managed to change the color of the girl’s dress.

How awesome is that?

So, looking back at all of the attempts…

What have we learned?

• Midjourney has difficulty assigning many specific details to different characters.
• This includes facial features, clothes, and especially color.
• Color bleeding is a recurring problem.
• Animals do not mix well with animals.
• Humans of the same gender also confuse Midjourney.
• Humans of different genders can work.
• In other words, multiple characters require strict archetypical contrast.
• Image references can be helpful for style, but only if your core text prompt already works on a standalone basis.
• Simple characters are easier because abstraction tricks our brains into believing that they look similar.

Frequently Asked Questions (FAQ)

Before we close off this guide, let’s quickly address some of the most common questions related to placing multiple consistent characters in Midjourney.

Conclusion

As you can see, despite being able to add image references that help maintain a consistent style, your core text prompts still need to be really good.

And that’s where multiprompts come into play because they can help reinforce concepts that otherwise might be misinterpreted if you phrase them incorrectly.

That magic lies within the clever combination of both techniques.

Here at Tokenized, we want to help you learn as much as possible about the AI software industry. We help you navigate the world of tech and the digitalization of our society at large, including the tokenization of assets and services.