Midjourney has a new feature that will make it even more difficult to spot fake news and ridiculous AI creations. It’s called “Character Reference,” it lets you take the character from one image and put it in many different scenarios, keeping it consistent throughout the images.

“This is similar to the ‘Style Reference’ feature,” Midjourney CEO David Holz writes on Discord, “except instead of matching a reference style it tries to make the character match a ‘Character Reference’ image.”

How character Reference works

First, create your character and upscale the image you like best. Then, type /imagine, your prompt, and type –cref with a URL to an image of the character.

Holz adds that you can use –cw to modify reference ‘strength’ from 100 to 0:

• strength 100 (–cw 100) is default and uses the face, hair, and clothes
• At strength 0 (–cw 0) it’ll just focus on face (good for changing outfits / hair etc)
• Cref works for both Niji and normal MJ models and also can be combined with –sref
• You can use more than one URL to blend the information /characters from multiple images like this –cref URL1 URL2 (this is similar to multiple image or style prompts)

Midjourney CEO moves on to explain how the feature works on the web alpha:

Drag or paste an image into the imagine bar, it now has three icons. selecting these sets whether it is an image prompt, a style reference, or a character reference. Shift+select an option to use an image for multiple categories.

What it’s meant for

“This feature works best when using characters made from Midjourney images,” Holz continues. In other words, it’s not designed for photos of real people, and it will “likely distort them as regular image prompts do.” Cref works similarly to regular image prompts except it “focuses” on the character traits.

However, there are some limitations. Holz points out that Cref won’t copy exact features like dimples, freckles, or T-shirt logos. From my really quick test, it doesn’t keep the character that consistent, and it introduces some weird traits and body poses. However, it’s still in alpha so I believe it will improve.

A quick test

I had to keep it simple because of time limitations, but I hope to play more with Character Reference over the next few days. I created this female photographer taking photos in nature and upscaled this image:

Then, I used Cref to make her pose for another female photographer… The scene didn’t quite go as planned, but the character traits are relatively consistent:

Then, I took the same character and made her dance in the street (while the Bowie x Jagger duet played in my head). This is where things got a little weird:

As I mentioned, the feature is still in alpha, so it can only improve. Holz points it out himself at the end of the announcement, too, noting that the Midjourney V6 official beta is coming soon. So, it’s time to have some fun with AI-generated portraits and narratives… But also to become extra careful and critical of the stories and images you see online!

[via PetaPixel]