For one of my recent projects, I needed to generate several illustrated character images. Each character needed both a sad and happy expression. While I could have used Midjourney’s character reference feature to generate consistent character images, I decided to try something different. This time, I used Midjourney’s new editor to generate additional expressions. Using the Midjourney editor meant I kept the same pose but just changed the expression. That worked better for this particular activity, where I was focused on depicting a transformation from sad to happy in the characters.
Image generation and editing process
Here’s the basic process I used:
- Generate the initial character image with one expression or emotion.
- In the Editor, use the Erase brush to remove the eyes, mouth, and eyebrows if needed to change the expression.
- Adjust the prompt to describe the new expression or emotion you want.
- Select the best edited image.
- Adjust and edit if needed. Sometimes you need multiple edits and prompts.
In the online activity, the characters fade from sad to happy expressions. Because most of the features stay the same, this makes the resulting transition very smooth.
Using the Midjourney editor
In the web interface for Midjourney, you can access the new editor on the right when you view a single image.
In the basic editor, the erase brush is active by default. You can brush over the parts of the image that you want to change. In this case, I want to change the eyes to be open and the mouth to be smiling. Those are the sections I erase from the image (which creates a very freaky image!).
I also changed the prompt to describe what I want to change about the expression.
Original prompt:
Vector illustration, a young girl feeling sad, head and shoulders, looking downRevised prompt:
Vector illustration, a young girl feeling happy, head and shoulders, eyes open, slight smileProblems with editing
Actually, when I originally tried this I didn’t explicitly say that her eyes should be open. That meant I got the smile fine, but all of the images kept her eyes closed. Even with “eyes open” in the prompt, one of the four choices had her eyes closed. In the image below, the bottom row shows four versions of the source image that I generated to change the aspect ratio from an earlier generation. The middle row is my first edit, without specifying that her eyes should be open. The top row is the eyes open version.
As you can also see in the top row, the eyes aren’t always proportional. That was a problem in many of my generations of revisions; the eyes weren’t consistent or didn’t quite match the original image.
It worked well enough when I only needed two expressions for each character. I think if I had needed 4 or 5 expressions for each character that it would have taken more time–possibly so much time that I would have been better off editing at least some of the images in another tool besides Midjourney.
Illustrated characters work better than photorealistic characters
This approach with the Midjourney editor works better with illustrated characters than with photorealistic characters. For this project, I used illustrated characters with a fairly simple style, which made the editing much easier. With photorealistic characters, I find it really hard to get the person to look the same. That’s true even when I use the –cref function to create a consistent character based on a reference.
For example, here’s a photorealistic character with a frustrated expression that I generated previously, plus four edited versions with slight smiles. Some of these might be OK as a match if you didn’t look at them too closely, but none of them are great. They resemble contestants in a look-alike contest rather than the original character.
Midjourney for images for elearning
While this isn’t perfect, and it sometimes requires multiple attempts, Midjourney is already a viable tool for creating images for elearning. I’m finding that in most situations I can generate images in Midjourney faster than I can find stock images.
That’s especially true for character images and when I need multiple specific images in the same style. Getting multiple images in the same illustration style has always been a problem in stock images; it’s one of the reasons I have tended to use photographs instead of illustrations. But Midjourney has changed that process for me.
Plus, if I spend less time searching stock image sites, I can spend more time on better writing or other activities to improve the quality of my work.
More on Midjourney
Looking for more on using Midjourney to generate images for elearning? Check out my previous posts.
“A cartoon character with a surprised expression, wide eyes, and raised eyebrows, set against a city street background.”
I tried your prompt, Hiru, and this is what I got.
Surprised character
You had good description of the facial expression you wanted, but not enough description of the character itself. Your prompt might work better with some additional details. For example, these two prompts are going to get very different results, even though most of the language is identical.
Prompt 1: “A cartoon character, white man, brown curly hair, red shirt, with a surprised expression, wide eyes, and raised eyebrows, set against a city street background.”
Prompt 2: “A cartoon character, Japanese woman, long black hair, red shirt, with a surprised expression, wide eyes, and raised eyebrows, set against a city street background.”