How to create consistent characters in Midjourney with this breakthrough method — the easiest yet
Use this Vary (Region) technique to create consistent characters that you can modify in any way you choose using the VariPose Connect Method
I went from this initial image:
To these images, keeping the character consistent, without face swapping, character sheets, image references, seed numbers or split screens:
Creating characters that remain consistent across various scenes is a common challenge for Midjourney users. Whether for children’s books, AI-generated videos, or other creative projects, maintaining a uniform appearance is crucial but often elusive.
There’s a good synopsis of current methods for achieving this goal of creating consistent characters in this article by Geeky Animals.
My method is closely associated with those methods, like the Split Screen method that Geeky Animals refers to in his article and is also similar to the Vary (Region) method employed by Morten Staveley, but I take it one step further (actually maybe several steps). I’m calling it the VariPose Connect Method. It allows you to put your character into various poses and settings and even allows you to interact with other characters.
There is no need to use split screens, character sheets, seeds or reference images. I do use “different poses, expressions” in my initial prompt but even that is unnecessary as you can insert a prompt for exactly what you want, whether it’s a particular pose, expression, clothing, pets and other objects you want to include in your scene. You can even imagine them in specific environments like the different job scenarios at the top.
You can follow my method on my YouTube video here, otherwise scroll down for a step-by-step guide:
The first step is to ensure that you have enabled “Remix” in your settings (if you’re using Midjourney on a Discord server):
Select the Remix mode button. This allows you to make changes to your prompt when you make changes to your images, after the initial prompt.
If you use the Midjourney site to create your images you don’t need to worry about this setting as it will be enabled automatically.
Experiment 1: Sonya (change the colour and style of her clothing)
I started with the following prompt based on the method that Geeky Animals used:
a portrait photograph of Sonya, isolated on blank white background, dynamic action | 2 split screen sequences --ar 16:9 --v 6.0
- and selected the following image from the four choices:
I then clicked on the Vary (Region) button…
and painted over the image on the right:
I changed the prompt so that she would now wear a pink top and pink sunglasses.
a portrait photograph of Sonya wearing a bright pink top with pink sunglasses, isolated on blank white background, dynamic action | 2 split screen sequences --ar 16:9 --v 6.0
This resulted in the following (selected) image:
From there I replaced the image on the left by repeating the Vary (Region) method above and painted over the image on the left:
I repeated a few times and ended up with the images below:
One last step — I removed the duplicate image on the left by painting over it using the Vary (Region) button and adjusted the prompt as follows:
blank white background --v 6.0 --ar 16:9
It might take a couple of tries to completely remove the other image and this is why it’s important to achieve a white background in your initial prompt. You should end up with an image like this:
From here you can pan up to bring the rest of her head into view and then use the Custom Zoom button to change the aspect ratio to your preferred size.
You may find that the character’s consistency may drift away from the original as you make more changes, so return to the source image if this happens and try again.
Basic structure for the initial prompt
By experimenting I’ve made further improvements on this method and discovered that using “portrait photograph”, “dynamic action” and “split screen sequences” are not necessary.
Including “isolated blank white background” and aspect ratio equal to 3:1 is my preferred method to generate at least 3 images that can be modified easily and where the background can be swapped out without difficulty.
In this next experiment I used a slightly modified prompt to the first prompt above but all you really need is the following prompt structure:
3 different images of [character name], [age] + [ethnicity] + [gender] + [hair] + [eyes] + [emotion] + [pose], isolated on blank white background --ar 3:1 --v 6 --style raw [--sref +url]
The parameters in square brackets are all optional. I use the character name to make it easier to search for that character again. The other parameters can be used to more precisely define the character, their emotion, their pose and anything else you require.
For example:
3 different images of Jessica, a 20-year old Hawaiian female surfer with long brown hair, bright green eyes and freckles, isolated on blank white background --ar 3:1 --v 6.0 --style raw
Optional: Style references for Version 6.0 and Niji 6.0
The style reference, which uses the --sref parameter and an image URL is useful if there’s a specific aesthetic you want to emulate.
If there’s a style you have in mind but you’re not sure how to describe it then look at the images that other people have created on Midjourney. If you find something you like that is close to what you had in mind then use the--sref style.
You can read more about using Style references on the Midjourney site:
Experiment 2: Katya (change her clothing and her environment)
I started with this prompt:
3 images of Katya, a 25-year old woman with blond dreads, bright green eyes and a wide smile, isolated on blank white background, variety of emotions, poses | split screen --ar 3:1 --v 6.0 --style raw
And managed to change her clothing, added sunglasses, furniture and even a pet dog after a series of prompt modifications:
To get from the initial prompt to the final prompt I applied the Vary (Region) button and selected the second two images from the initial prompt, keeping the first image. It’s important to always keep at least one image to ensure consistency.
I changed her into a pink top with blue sunglasses by changing the prompt:
3 images of Katya, a 25-year old woman with blond dreads, bright green eyes and a wide smile wearing a pink top and blue sunglasses, isolated on blank white background, variety of emotions, poses | split screen --ar 3:1 --v 6.0 --style raw
Then I panned down and used custom zoom until I reached this point:
I then painted over the left-hand image using:blank white background --v 6.0 --style raw --ar 25:21
I next changed the aspect ratio to 16:9 by using the Custom Zoom button and adding--ar 16:9 --zoom 1, which resulted in the following image:
From there I painted the entire area to the left of the woman using the following prompt:
Print of vibrant beach scene hanging on the wall with rattan furniture underneath --v 6.0 --style raw --ar 16:9
After those changes I ended up with this image:
By selecting the entire area to the right of the woman and adding the following prompt I ended up with the final image:
Large Dalmatian sits next to Katya --v 6.0 --style raw --ar 16:9
As you can see this method is quite powerful but you need to get the initial prompt right to start off. You can specify the environment in your initial prompt but it can be more difficult to make changes to your character afterwards.
Experiment 3: Jessica (change facial features, emotions, environments)
I started with this prompt:
3 different images of Jessica, a 20-year old Hawaiian female surfer with long brown hair, bright green eyes and freckles, isolated on blank white background --ar 3:1 --v 6.0 --style raw
And I modified the prompt (in bold):
3 different images of Jessica, a 20-year old Hawaiian female surfer with long brown hair, bright green eyes and freckles wearing a green cap and light blue sunglasses, isolated on blank white background --no vignetting --ar 16:9 --style raw
The result looked like this:
I used the Vary (Region) method to create the following variations:
Heterochromia (different coloured eyes), change to blonde hair,
I also experimented with different emotions. This was difficult to achieve, I used ChatGPT to help with describing the physical signs of the emotions I wanted to emulate. In this case: surprise, sadness and anger.
Finally, I used ChatGPT to suggest different jobs that Jessica might take on. Here are a few of the results:
Can you use this with the Niji (Anime style) model?
Yes. See some of the results below.
I hope you found this article useful. Midjourney is due to roll out their new character consistency solution soon so this article may soon become redundant.
The--cref parameter is expected to allow creators to change colours and backgrounds whilst keeping the character consistent. We’ll have to wait to see how that works in practice. Hopefully it will make the task of creating consistent characters a lot easier. In the meantime, have fun with this Vary (Region) method.
