You’ve probably come across viral posts featuring incredibly consistent AI-generated characters.
But, just like me, when you try to create similar characters using the Midjourney AI bot, Starry.AI, or Bing Image Generator, you just can’t generate the same person more than once.
Your outputs might come out looking something like these, eerily similar, yet totally different.
That’s how my attempts at creating consistent characters went, until recently. But now I’ve cracked the code!
I’ll show you how to create consistent characters to use for your art projects, video games, website pictures, or book covers using the Midjourney AI bot that look a bit like this:
Don’t worry! The principles in this tutorial will work for cartoons, anime, and any other style or subject you want to repeat across AI generations.
AI Image Generators: Midjourney
The Midjourney AI bot is one of the leading AI text-to-image generators on the market, and the one I like to use. It takes an input text prompt, eg. “A photo of an apple”, and turns it into an AI generated image.
Midjourney is responsible for most of the viral AI images, including this one of the pope wearing a drippy outfit.
It’s a paid service, so check their pricing page here. If you need help setting up Midjourney and using it with Discord, follow this straightforward tutorial.
Step 1: Create a Base Character
Midjourney Version 5 excels at generating photorealistic images, but you can choose any style or subject.
If you’re looking for prompts to create anime, cartoon, photorealistic, or stylized characters, head over to the Promptlib homepage. There are thousands of prompts you can easily copy and paste directly into your Discord window to generate any character or style.
My base character:
PROMPT
half body portrait of a blonde woman holding a screwdriver, mechanic outfit, mechanic shop background, shot on kodak gold 400
Step 2: Copy Your Image Link
This is crucial. Instead of using the Midjourney AI bot’s built-in seed feature, copy the image link and use your base character as a source image.
Not only do the images turn out better, but you can influence them with image weights to strengthen the influence of your source image.
This article explains how and image weights can enhance your source image’s effect on the output:
To use your picture as a source image, right-click and select “Copy Link”, or “Copy Media Link”.
Add this link to the beginning of your text prompt, and the resulting generation will be heavily influenced by the source image.
Creating Different Scenes
Your text prompt is vital, as it influences the photo angle, pose, and other factors. Consider the scene you want to create, but remember that certain scenes can be challenging to achieve depending on the source image.
For example, my base character held a metal pipe, so even when I tried to place her in a dress, Midjourney kept generating the pipe:
Midjourney tends to struggle to alter a scene without also changing the features of your base characters.
However, there are some helpful tips that can help to minimize these limitations. I’ve found the following techniques to be effective.
Getting the Desired Scenes, Poses, Angles, and Results
- Crop out the face: To resolve the pipe issue, I cropped out her face and used it as a reference image. This gave the AI less information about the body, but it allowed more flexibility with angles and poses, and the pipe was gone!
As you can see below, this made a massive difference. In the first picture the pipe is front and center, and in the second it’s gone.
Generation using uncropped source
Generation using cropped source
- Use photography terms to alter the scene: Include terms like “action photography” or “wide shot” in your prompt to control what type of picture you generate, learn more about writing prompts for AI images and how to achieve the output you want.
Action photography of a blonde women laughing
Full shot of a blonde woman in a dress
- Be realistic with your prompt: The Midjourney AI bot can produce nonsensical images if you include too many details or subjects. Keep it simple. The longer your prompt, the less Midjourney will reference your original picture, and the less similarity it will have.
PROMPT
action photography of a blonde women by a broken down car with neon lights, sparkly outfit, a man is beside her with black hair, at nighttime, neon lights shot on kodak gold
- More is more: You’ll need to generate a whole lot of pictures to get 5 good ones, around 10x times that number. I made dozens of almost-dopplegangers in the process of generating my set of cohesive pictures.
By implementing these tips, I was able to generate way more consistent characters, as you can see below.
What do you think?
While not perfect, this is currently the closest Midjourney can achieve in terms of consistency. We’d love to hear your thoughts on this tutorial and see your consistent characters in the comments. Feel free to share any feedback or questions about where you got stuck!
Recommended from Promptlib
Create a Personalized Children’s Book – In Just Hours!...
Make a Custom Children’s Book With AI ....
Read more.
Everything You Need to Know About GPT-4: Explained...
After months of anticipation, Microsoft-backed startup OpenAI ....
Read more.
Artificial Intelligence Won’t Steal Our Jobs, It’ll Change Them...
With AI’s capability to complete generalized tasks ....
Read more.
Creating Characters with AI: Midjourney Guide for Artists, Writers, and Game Designers
Everything You Need to Know About GPT-4: Explained
Robot Assistants: Is Microsoft’s ChatGPT Turning Sci-Fi into Reality?
Seamless Tiles, The First Real Use Case for AI-Generated Art?
