How To Create Consistent Characters in Midjourney

Effective storytelling often depends on the use of consistent characters, with the same appearance and style from one image to the next. Maintaining character consistency across different scenes allows individuals to track them effortlessly, understand their unique traits and behaviors, and connect with them.

AI art tools like Midjourney, DALL-E, and Stable Diffusion can produce different characters to inspire your creativity. But getting these systems to generate the same characters in various settings, poses, and moods consistently can be a challenge. This is because AI-generated images are usually random and don't reference previous artworks.

But with the right strategies and techniques, an AI art generator such as Midjourney can help you create consistent characters that you can use in different media, including books and comics.

Keep reading to discover how you can use Midjourney to create consistent characters.

Midjourney basics

Midjourney is a generative AI tool that's tailored specifically for image generation. Based on your prompts, Midjourney can produce images in different styles, including photorealistic, surrealistic, minimalist, and mosaic.

With its image generation capabilities, Midjourney can help you generate characters for your story. Such consistent characters facilitate better narrative coherence, recognition, and cross-media adaptation.

To follow this step-by-step tutorial, you must have some basic experience with Midjourney. Particularly, you should know your way around the Midjourney Discord server and understand how to submit new image generation prompts and how to evaluate the outputs.

New to Midjourney? Learn the fundamentals of generating images.

‍Note: Midjourney is a paid service. So, you'll need to be a paid subscriber to access different features like fast image generation. In this tutorial, we'll be using the latest version, Midjourney v7.

Let's dive in.

4 steps for creating consistent characters in Midjourney

Consistent characters are essential in storytelling, particularly in digital media, children's books, and comic books. This is because they help maintain narrative cohesion. In other words, having consistent characters allows the audience to easily understand what you're trying to convey as the plot moves along.

For example, in our case, we'd like to develop a simple character for a children's book. This character should be in different poses and moods to fit in various scenes to tell our story.

Here are steps to follow when generating consistent characters in Midjourney.

1. Set up your initial character reference
2. Create a character sheet
3. Craft your base prompt
4. Maintain consistency across prompts

1. Set up your initial character reference

The first step is to create an original image of your character that you'll use as a style reference (sref). This image will be the foundation for all future iterations, helping you maintain consistent character features.

While you can sketch the initial character manually, a good alternative would be to generate the character in Midjourney. 

We used the prompt, "portrait of a young woman, big expressive teal eyes, long wavy auburn hair, soft rounded face, natural smile, light freckles, wearing a detailed green embroidered dress, Pixar style, clean lighting, white background --ar 2:3 --v 7.0" to produce the following character. 

The --ar extension in the end of the prompt allows us to specify the aspect ratio that Midjourney should use and the --v is the MidJourney version.

MidJourney will generate four different character variants, of which you can choose the best one:

In our case, we chose the first character.

2. Create a character sheet

After generating the initial character, you can now create a character sheet using this reference image. This sheet will contain multiple poses and expressions of the character, which you'll use for different scenes in your story.

To get started, download the image from Midjourney onto your computer. If need be, use image editing software like Photoshop to crop the image to focus more on the character. In our case, we don't need to crop since the character is already in focus (portrait mode). 

We can then upload the image to a Midjourney channel and copy the image address or URL.

Next, click the uploaded image to access an expanded view, and then right-click on it to copy the image address (URL).

‍In our case, the URL is:

https://media.discordapp.net/attachments/941971306004504638/1412768815124709416/wanjamike_portrait_of_a_young_woman_big_expressive_teal_eyes_lo_4a475fd6-dee3-4a4a-916f-fb2ba316319e_1.webp?ex=68b97efd&is=68b82d7d&hm=158421f14e7704cd836522dbe59e889c984ac059f21a7609be2f7894965d485b&=&format=webp 

We can then add the above image URL (Midjourney will use the image as a reference) to our prompt as follows.

Note: We're using the same original prompt we used to generate the image, but in this case, we're adding extra details like multiple poses and expressions and a character sheet.

‍

Here are the four character sheets produced from the above prompt. Though the generated character on the sheet does not fully match our initial character, some elements, like hair, dress, and facial expressions, are similar. 

Note: We used the - -oref to reference the initial character we generated in the first step. Currently, Midjourney v7 does not support the - -cref annotation.

Click on the AI character sheet you like best to expand it. In our case, we liked the fourth character sheet because it closely resembles our original character. Here's how it appears.

3. Craft your base prompt

Before crafting a base prompt, crop each character in the above character sheet into a separate image, as shown below. It's a good idea to upscale your character sheet to increase the size and resolution of your character in different poses.

We can then add the images to the Midjourney Discord channel and copy the links.

While our AI character sheet contains six images of the character, we don't have to include all of these variations in the base prompt. In this case, we'll reference two images — of the character in varying poses — to allow Midjourney to generate the desired outputs. These are the image links for the above characters:

‍

To create the base Midjourney prompt, we need to add the above links to our command and then specify what the character should be doing. In this example, we want the character to appear to be laughing. We will use one of the links as the omnireference (oref). 

Since Midjourney may misinterpret laughing as merely smiling, we have improved the prompt with "big open-mouth laugh with visible teeth." Here's our original prompt with the added descriptor. 

‍

Midjourney generated the following image variant of the character laughing.

Building on the first image, we created another one to emphasize the character's warmth and affection by having her hands form a love symbol.

We proceeded to change the setting of the character using the following short prompt.

As with many AI-generated images, check carefully for visual anomalies — especially with hands, facial symmetry, or interactions between features like eyes and cheeks. If you spot something off, try rerunning the prompt with slight wording changes or use the Vary Region tool in Remix Mode to fix specific problem areas.

‍

You may have to run the text prompt multiple times to get your desired images, tweaking various parameters as you go. Though the images may have flaws, the character's facial appearance is consistently similar. This helps set the stage for a consistent character design.

Tip: Tools like OpenAI's ChatGPT can help you craft richer, more detailed image prompts for your Midjourney characters.

4. Maintain consistency across prompts

You should also include the –seed parameter on future prompts to boost character consistency. This parameter helps in maintaining the character's recognizable features across various depictions.

To generate a seed number, simply click on a generated image and then react with an envelope emoji.

The Midjourney bot will send a seed number to your Discord inbox, as shown below.

‍We can then add the seed number to a new prompt.

‍

Here's the Midjourney output of the character with a sad face.

Here is an upscaled version of the first image in the above character sheet:

We used this prompt to place the character near a river:

‍

‍Note: This link is a shortened URL automatically generated by Midjourney when you upload or reference an image in Discord. You can copy it by clicking on the uploaded image and selecting "Copy link."

Strategies for creating consistent characters

Creating consistent characters in Midjourney can be challenging, but it can also be a thrilling experience. Here are some strategies to help you generate consistent images.

1. Describe facial features clearly

While an AI image generator like Midjourney can produce incredible visuals, you have to use detailed text descriptions to get the desired outputs. Specifically, use clear and consistent descriptions of your characters' facial features to produce more accurate images. These facial features will allow the characters to portray different moods and expressions as the story flows effectively.

In the images generated in the previous section, we gave Midjourney more creative freedom by only specifying that the character should have big expressive teal eyes, long wavy auburn hair, and a soft rounded face. However, you can use phrases like "eyes wide open, eyes narrowed, a mustache, and smile on the face" to create more descriptive prompts.

2. Avoid conflicting keywords

Remember to use consistent language when generating images in Midjourney. Avoid conflicting keywords, which may lead to inconsistent results.

For instance, defining a character as tall and then describing them as short afterward may confuse the model. Using clear, simple, and short phrases can help you maintain consistency when generating multiple characters.

3. Use the seed number

Using an image's seed number might also help you generate consistent character variations in Midjourney.

To illustrate, we used the prompt "generate a photorealistic picture of a middle aged man, portrait mode, white background --ar 2:3 --v 7.0" to produce the following images.

We then retrieved the seed number from one of the characters:

We then used the same prompt, but adding the seed number, to generate the following nearly similar characters.

But if we retry the same prompt but without the seed number, the outputs are quite different:

‍Once again, the seed command may not automatically result in consistent images. So, tweak your prompts to improve the likelihood of similar outputs.

Note: When working with photorealistic prompts, it's a good idea to reverse image search your results to ensure they don't unintentionally resemble real people — especially public figures. If a generated character appears too close to a known individual, consider adjusting your prompt descriptors and rerunning the generation.

4. Use the Vary Region method

You can also use the Vary Region feature to edit specific sections of an image while maintaining the rest of the character integrity for better consistency.

While we can use the standalone Vary Region feature, combining it with Remix Mode allows us to submit new or edited prompts to modify specific image portions. This can lead to outputs that are much closer to what we want.

To get started, first enable Remix Mode using the command "/prefer remix." Type this directly into any Midjourney Discord channel where you generate images. You should see a confirmation message once Remix Mode is turned on.

Next, proceed to upscale an image on a grid and then press the Vary Region tool to access the Midjourney image editor.

To illustrate, we will use one of the characters we generated in the above step.

With Remix Mode turned on, the Midjourney editor will look as shown below.

You can then mark the areas you want to edit (click and drag your cursor over the part of the image you want to change to highlight the editable region). In our case, we want the character to have a hat. 

We also modified the prompt to read as:

‍

Here is the Midjourney output:

You can turn off Remix Mode using the /settings command and deselect the Remix mode button, shown below.

‍

‍5. Use character sheets and style guides

While creating characters, it's a good idea to generate a character sheet that shows different angles and expressions, as we did with the red-haired girl. These character sheets ensure consistency across various depictions, such as comics and anime.

A character sheet also shows how characters appear in different poses and expressions. As a result, artists can use this sheet to modify characters to ensure they accurately bring out their personalities and portray specific emotions.

The prompt "a short girl, long hair, white t-shirt, blue pants, multiple poses and expressions, character sheet, Pixar style --ar 16:9" generated the four following character sheets in Midjourney. Again, we'd select the set we like or tweak the prompt until we get the best match for our needs.

Get help with Midjourney on Upwork

Whether you're designing characters for a children's book, anime, or storyboards, Midjourney can help with your character creation process. By referencing other images, you enable Midjourney to generate characters that closely resemble one another.

However, drafting the right prompts to get Midjourney to produce consistent characters requires time, especially if you're a beginner. While Midjourney can create characters with consistent facial features or those that resemble real people, getting the AI tool to match clothing or other elements of the image can be challenging. Consider working with Midjourney AI artists on Upwork to help you harness the power of AI in your creative workflow.

If you're an expert looking for work, Upwork can connect you to different Midjourney AI jobs to help build your portfolio. Get started today!

FAQ

We have answers to common questions about using Midjourney to create consistent characters. 

Can I make Midjourney remember my character automatically?

No, Midjourney doesn't keep copies of past generations. To keep your character consistent, you'll need to use reference images, seed numbers, or character sheets in your prompts.

How do I stop Midjourney from changing my character's clothes?

Midjourney often tweaks details like clothing unless you lock them in with clear, consistent prompt descriptions. As of Midjourney version 6, you can now use the character weight parameter (--cw) to help maintain consistency across generations, which reduces unwanted outfit changes. For fine-tuning, the Vary Region feature in Remix Mode can let you edit just the clothing while keeping the rest of the character intact.

Do I need Photoshop to create a character sheet?

Not necessarily. Photoshop or similar photo editing tools help crop and organize character poses, but you can also rely on Midjourney alone to generate multiple angles and expressions on a single sheet.

What's the fastest way to generate consistent characters for a storybook?

The quickest method is to create an initial reference character, then generate a character sheet with multiple expressions, and reuse image links and a seed number in all prompts. This workflow minimizes random variations and speeds up consistent output.

‍Upwork is not affiliated with and does not sponsor or endorse any of the tools or services discussed in this article. These tools and services are provided only as potential options, and each reader and company should take the time needed to adequately analyze and determine the tools or services that would best fit their specific needs and situation.