How do I create consistent characters in Midjourney?

Within two years, image AI has radically changed the way we work with images and graphics. Until recently, we were either reliant on boring and repetitive stock images or had to pay for expensive photo shoots if we wanted to enrich our web and social media posts with visual elements. However, today tools like Midjourney or Dall-E allow us to realize our visual ideas in seconds.

I wrote several months ago about how rapidly the quality of images generated by Midjourney has improved.

However, one problem remained despite all updates: it was only with difficulty possible to generate consistent characters. Yet, this feature is indispensable if one wants to use Midjourney, for example, to create comics or illustrations with the same character.

Personally, I run a history YouTube channel that introduces historical figures. Until now, I have always faced the problem that the characters looked different in every scene.

But now, this difficulty also belongs to the past, because with the latest Midjourney update, the tool finally allows the creation of consistent characters.

How do I create consistent characters in Midjourney?

Within two years, image AI has radically changed the way we work with images and graphics. Until recently, we had to rely on boring and repetitive stock images or pay for expensive photo shoots if we wanted to enrich our web and social media posts with visual elements. Today, however, tools like Midjourney or Dall-E allow us to realize our visual ideas in seconds.

I wrote several months ago about how rapidly the quality of images generated in Midjourney has improved.

However, one problem remained despite all updates: It was only with difficulty possible to generate consistent characters. This feature is, however, indispensable if one wants to use Midjourney to create, for example, comics or illustrations with the same character.

I personally run a history YouTube channel that introduces historical figures. Until now, I have always faced the problem that the characters looked different in every scene.

But now this difficulty also belongs to the past, because with the latest Midjourney update, the tool finally allows the creation of consistent characters.

To use the feature, I have to upload a reference image in Midjourney and add the command —cref followed by the link to the image within a bracket to the prompt.

The finished prompt looks like this:

/imagine Prompt Text –cref {https://linktext}

With the command –cw 0-100, I can also specify the level of detail to be adopted. —cw 100 would adopt all details of the character down to the clothing, while —cw 0 would only adopt the face.

This is useful for creating characters with different clothing styles.

Consistent characters can be created in Midjourney Alpha or traditionally in Discord.

In Discord, the image must be uploaded and then opened in a new tab by right-clicking on the image and selecting “Open image in new tab”. Now the URL can be copied and inserted into the prompt.

The process is easier in Midjourney Alpha. Here, simply upload the image and then select the option “Use as Character Ref” from the hover menu (Alternatively, the options “Use as Style Ref” and “Use as Image” are available).

How well does creating consistent characters work in Midjourney?

As we can see, it is very easy to work with the —cref and —cw commands. But how good are the results?

To answer this question, I have been intensively experimenting with this function over the last few days. I started with my personal profile picture. My goal was to see if I could easily place myself in new situations using this picture. This could allow me, for example, to create new images for my website without having to book a lengthy and expensive photo shoot.

I chose the following picture as a reference photo:

From this I generated the following images (you can find the prompts in the image description):

How good do I personally like the results? The generated midjourney character definitely looks different than I did in the original reference photo. Nevertheless, the character is relatively consistent across the different images.

What happens if I use multiple images of myself as reference images?

In this case, we just need to upload several images to Midjourney Alpha and mark them as character references.

And here is the result:

That’s still not me, but this version looks a lot more like me.

We see: By uploading several reference images, I can achieve significantly better results, even if the results achieved are not yet perfect.

By the way, Midjourney’s recommendation is that this feature works best with images generated by Midjourney.

I tried that too. As a reference photo I took this young woman from the Midjourney explore-gallery:

I used her to create the following images:

Here, too, the original image was not 100% adopted. The results are good enough to create consistent characters for illustrations or comics, although of course you are less critical with other people’s characters than with your own images.

Conclusion: Consistent characters in Midjourney

The function doesn’t work perfectly yet, but it works well enough to be used for simple projects such as my YouTube channel. Currently the official recommendation is to use this feature with images generated by Midjourney, but as I’ve shown the results are relatively consistent even with images of real people, they just don’t reflect the person 100%.

What is clear, however, is that we are less critical of generated people than we are of images of real people or ourselves. The results also became significantly better when I used several reference photos.

Looking at Midjourney’s impressive development since 2022, I am confident that this feature will soon deliver significantly improved results.