How to create consistent characters in Midjourney
A guide for generating the orthographic front and back views of AI-created characters
Creating character art for game development can be a challenging and time-consuming process, especially for independent game developers who may not have access to the same resources as larger studios. However, with the help of tools like Midjourney, creating character art has never been easier or more accessible.
In this article, we are going to use Midjourney AI to create consistent-looking character art for game development. We are going to show you step by step how you can create the same character in different poses and actions, plus, amongst all generate orthographic front and back views. Keep in mind results aren’t perfect yet, however, they can be incredibly useful.
What is Midjourney?
Midjourney is an independent research laboratory that has developed artificial intelligence software that generates images based on textual descriptions (prompts). With Midjourney, game developers can create highly detailed and lifelike character art in a fraction of the time it would take to create manually.
How to start creating in Midjourney?
To start generating with Midjourney you first need to sign up with an account. Once you have signed up you have to join their Discord server and accept their terms of service.
Once you are onboard you can go to a newbies channel and start creating. To start generating art you have to type /imagine [prompt] — inside this prompt is where you describe what you want the AI to generate. You can be as specific or vague as you would like.
eg. /imagine[an African warrior from the future in full armour]
Creating a character with multiple views
Firstly, in order to create a character you can decide between:
- a character sheet or
- the concept art of the character with the front and back view
Once you decide on the approach, this will be the first parameter of the prompt we type in (you can always generate the other approach afterwards).
Next, we need to visualise our characters and bring them to life by deciding on the name of the character. What does the character look like? Whenever you are creating a character, giving them a name will help the AI associate your current generation with future ones, as well as help you, as the artist, relate to the character. Subsequently, we want to write a description of what the character looks like with as much detail you want. You can describe the colour of the hair, any special features, age range, etc.
Once your character description is written, we need to specify the view of the character we are looking for.
Here we would suggest writing:
‘full body view, orthographic full view of [Character name] facing the camera, orthographic full view of [Character name] facing away from the camera’.
This will specify that the generated character should be in full view (head to toe) and that we require an orthographic front and back view of said character. Keep in mind that most characters will be generated in an ‘A’ pose by default, however, this is not guaranteed. If you require an ‘A’ pose you can specify in the command by adding “arms stretched to the side”.
eg. ‘full body view, arms stretched to the side, orthographic full view of [Character name] facing the camera, orthographic full view of [Character name] facing away from the camera’.
Additionally, we want to specify to the AI the art style for your character. You can write things like “cartoon”, “hyperrealistic”, or “drawn with crayons”.
Optionally, we suggest specifying to Midjourney that your generations should be created with an aspect ratio of 3:2 by adding in the end:
— ar 3:2
This will allow more space for your character views to be generated.
Once you click enter and send your request through to Midjourney, you will receive four image variations of the same prompt within 60 seconds.
Adding it all together, our recommended prompt template looks as follows:
/imagine [ Character sheet / Concept character art, Character name, Description of character, full body view, arms stretched to the side, orthographic full view of [Character name] facing the camera, orthographic full view of [character name] facing away from the camera, art style (eg. Cartoon, sketch, hyperrealistic, etc) — ar 3:2
Eg: /imagine [ Concept art character, Dale, an African warrior from the future in armour, full length body view, arms stretched to the side, orthographic full body view of Dale facing the camera, orthographic full body view of Dale facing away from the camera, hyper detailed, hyperrealistic — ar 3:2]
Once you are happy with one of the generations you received, we would suggest creating variations of the same by clicking on V# (note: the images are numbered 1–4 from left to right). This will generate 4 variations of the same generation.
Whenever you find a variation that you are happy with, you can proceed to upscale by clicking U#(note: the images are numbered 1–4 from left to right). Upscaling simply means upsizing the single generation you want.
How to maintain consistency?
Now that the image has been upscaled into a full-size, full-quality image, what you want to do in order to create more images of the same character in different actions/ poses, is open it in a browser, and copy the image address (right-click copy > image address). We do this because we’re going to use this image address as a reference for our next prompt to try to create the same character. Keep in mind the more references you can give the AI about this character, the better.
Additionally, you need to get your generation seed number.
What are seed numbers?
Seeds are basically reference numbers within Midjourney that will allow you to create the exact same or similar images from one prompt to another. This is going to be very useful for us as we want to create a very similar character, if not exactly the same from picture to picture, doing different things.
In order to get your seed number:
scroll up to your generation > click on Add Reaction > click on the envelope emoji ✉️
This will generate and send you the seed number through the Midjourney bot. Using the same seed number and prompt will produce similar ending images, which is exactly what we need.
Now that we have all three things (image address, prompt & seed number) we want to put it all together and generate more images of the same character doing something different. Here is how to go about it:
First, we paste the copied image address in the prompt, followed by your original prompt (with any changes you would like in action, pose, accessories, etc.), and finally “ — “ followed by “seed XYZ”.
eg. /imagine [https://cdn.discordapp.com/attachments/1055917701199036506/1077983589016272966/Kaedim_Concept_art_character_Dale_an_african_warrior__773d8853-2375-43cf-9640-e2e72d347d60.png, Concept art character, Dale, an African warrior from the future in armour and sword, full length body view, orthographic full body view of Dale facing the camera, orthographic full body view of Dale facing away from the camera, hyper detailed, hyperrealistic — ar 3:2 — seed 2911151543]
Once your image is generated you can make variations of it again and upscale it until you have your ideal character.
It is important to remember that all variations of the same character will not be perfect as this AI is still developing, however, it can save hours of work and can achieve incredible and unexpected results.
Now it is your turn to try it out! Let us know how it goes in the comments.
Want to accelerate your 3D workflow and ship 10x faster? Kaedim uses AI to convert 2D art to 3D content for game devs. Come talk to us!
