A Guide to Consistent Characters with –cref

Quick Links

• What is Midjourney’s New –cref Parameter?
• How to Use –cref
• Can You Use More Than One Link in –cref?
• Does –cref Work?
• Why Isn’t –cref Working?
• Can You Use –cref in Photography?
• How to Use –cw

What is Midjourney’s New –cref Parameter?

The wait is finally over!

Midjourney just released a new parameter that helps you generate Consistent Characters.

Now you should be able to generate your character doing different actions from different angles.

However, I will warn you this is a brand-new feature that has its limitations.

Related Articles:

• Lighting Prompts for Midjourney
• Generate Realistic Images in Midjourney

How to Use –cref

The cref parameter works very similarly to the –sref one.

Simply add the parameter to the end of your prompt and then add the URL of your character reference image.

How do you get the url of your image? I go over how to do this in my image prompt article!

Can You Use More than One Link in –cref?

You only need one image link for the parameter to work, but you can use as many as you’d like.

That being said, I didn’t find that adding multiple links to the –cref parameter helped when I tested it. Adding multiple –sref links , however, really helped the AI narrow in on your style. See my examples here.

Does –cref Work?

…kind of.

This is a brand new parameter that is built for a version of Midjourney that isn’t even out of Alpha testing yet, so it’s far from perfect.

However, it’s a great first step in allowing us designers to generate the start of a consistent character.

You can prompt for your character at different angles, feeling different emotions, or doing different tasks and get inspired by what the AI generates.

It is not an exact replica of your character yet, but you can see how the character is repositioned and then edit your character to match.

Read on to the next section to learn how to get around some of –cref’s initial limitations.

Why Isn’t —cref Working?

The –cref parameter is a new parameter based on a version of Midjourney that isn’t technically out of its testing phase yet. It’s going to be buggy. ¯\(ツ)/¯

Here are some limitations I ran into already:

It Only Works on V6

This parameter only works on v6 as of the writing of this article. This shouldn’t be a huge problem since v6 was recently made the default version for all users. However, if you have set your default version to v5.2 or lower in your settings, you will need to use the —v parameter or set v6 as your default in your settings.

You Need to Define the Style with –sref

The biggest flaw I’ve struggled with is how –cref gets the style of my characters SO. WRONG!

How can I create consistent characters when the AI is changing my flat vector illustration into a surrealist painting??

Check out what happens when I try –cref with this illustration I made of my husband for our custom AI chatbot business, Fulton St Digital.

Original photo

When I upload this image with just the —cref parameter and a simple prompt “thinking”, Midjourney takes some serious liberties with the style.

Just –cref

The resulting images look like 3D paintings, completely losing the original flat vector style! This is totally unacceptable if I am trying to get inspiration for how my character would look if he was thinking. It just sort of defeats the purpose of the parameter.

However, if you teach the AI your style with a –sref photo, you can generate more accurate photos! You can even use the same URL you used in the –cref parameter as your style reference.

With –sref

If you have more than one image you can use as a style reference, it can be helpful to reinforce the AI’s understanding of that style.

Additional Style References

Results

Try Using —style raw Parameter

Since Midjourney is adding its own style on top of the style reference you are providing, sometimes the AI can generate images that are slightly off of what you are looking for. Removing that default style with the —style parameter helps ensure the AI stays focused on the references you provided.

Increase the –sw

Although the –cw is at its highest value by default, you can increase the style weight further. The default –sw is 100, but the value can go all the way up to 1000. Increasing it can help Midjourney better understand your character’s style and, therefore, create a closer replication of your original image.

Do Not Use Image Prompts

Image prompts give Midjourney a reference of the type of composition you are looking to generate. So if your image prompt is of a character standing, but you prompt for a sitting character, 9 times out of 10 Midjourney will generate a standing character.

Essentially, this means it will generate a worse version of the photo you uploaded as an image prompt.

You can override this slightly with the —iw parameter, but you’ll need to set the value so low that you’d lose any benefit the image prompt provided anyway.

Can You use –cref for People in Photographs?

While Midjourney warns that –cref is not for “real photos”, I actually found the AI better at generating similar-looking models in photographs than it was at replicating illustrative characters

Original Photo

Photo After –cref

How to Use the —cw Parameter?

–cw is a special parameter built just for –cref. Similar to –iw and –sw, –cw controls how much weight you’d like to put on the provided character reference links.

You can weigh your references anywhere between 0-100.

The default weight is 100, which means the AI tries to replicate the character’s facial features, hairstyle, and clothing.

The closer to 0 you get, the more the AI will disregard the character’s clothing and hairstyle. It will always try to replicate the facial features even with a –cw of 0.

This is helpful if you are trying to generate your character in a different set of clothes.