Midjourney gen introduces permanent roles for AI images.

Popular AI image-generating service Midjourney has deployed one of its most requested features: the ability to continuously recreate characters in new images.

This has been a major obstacle for AI image generators by their very nature to date.

This is because most AI image generators rely on “diffusion models”, similar to or similar to Stability AI’s Stable Diffusion open-source image generation algorithm, that take text input by a user. And works roughly by trying to piece together an image pixel by pixel. -pixels that match this description, as learned from similar images and text tags in their large (and controversial) training dataset of millions of human-generated images.

Why are persistent characters so powerful—and elusive—for creative AI imagery?

Nevertheless, as is the case with large text-based language models (LLMs) such as OpenAI’s ChatGPT or Cohere’s new Command-R, the problem with all generative AI applications lies in the inconsistency of their responses: the AI ​​responds to each prompt. creates something new for is entered, even if the prompt is repeated or some of the same keywords are used.

It’s great for creating entirely new pieces of content—in Midjourney’s case, photos. But what if you’re storyboarding a movie, a novel, graphic novel or comic book, or any other visual medium you want? same as Does the character or characters move through it and appear in different scenes, settings, with different facial expressions and props?

This precise scenario, which is usually necessary for narrative continuity, has so far been very difficult to achieve with generative AI. But Midjourney is now taking a crack at it, introducing a new tag, “–cref” (short for “character reference”) that users can add to the end of their text prompts in the Midjourney Discord and the character’s name. Will try to meet face to face. Features, body type, and even clothing from the URL that the user pastes into the following tag.

As this feature develops and improves, it could take the Midgerni from being a cool toy or ideation source to a more professional tool.

How to use the new Midjourney persistent character feature

The tag works best with pre-made midjourney images. So, for example, the workflow for a user would be to first create or retrieve the URL of a previously created role.

Let’s start from scratch and say we’re creating a new character with this prompt: “A muscular bald man with a bead and an eye patch.”

We’ll upscale the image we like best, then control-click it in the Midjourney Discord server to find the “Copy Link” option.

Next, we can type a new prompt “wearing a white tuxedo standing in a villa -cref [URL]And paste in the URL of the image we just generated, and Midjourney will try to generate the same character as before in our newly typed layout.

As you’ll see, the results are pretty consistent with the original character (or even our original hints), but are certainly encouraging.

In addition, the user can somewhat control the “weight” of how closely the new image reproduces the original character by adding the tag “–cw” from 1 to 100 to the end of its new prompt (” After” -Cref [URL]” string, so like this: “-cref [URL] -cw 100. The lower the “cw” number, the more variation there is in the resulting image. The higher the “cw” number, the more closely the resulting new image will follow the original reference.

As you can see in our example, entering the minuscule “cw 8” actually returns what we wanted: the white tuxedo. Although he has now removed our character’s distinctive eyepatch.

Oh well, a little “different terrain” can’t fix it – right?

Okay, so the eye patch is on the wrong eye… but we’re getting there!

You can also concatenate multiple characters into one using two “–cref” tags along with their respective URLs.

The feature went live earlier this evening, but already artists and creators are testing it now. If you have migraines, try it yourself. And read founder David Holz’s full note about it below:

Hey @Everyone @ Here we are testing a new “Character Reference” feature today. It’s similar to the “Style Reference” feature, except instead of matching a reference style, it matches the character as a “Character Reference”. tries to match the image of

How it works

• Kind of --cref URL Following your prompt with the URL for a character image
• You can use --cw To change the reference ‘power’ from 100 to 0
• Power 100 (--cw 100) is the default and uses face, hair and clothes.
• 0 at power (--cw 0) it will focus only on the face (good for changing clothes/hair etc)

What does that mean?

• This feature works best when using characters created from Midjourney images. It’s not designed for real people/images (and will likely distort them like regular image notations do).
• Cref works like regular image notation except that it ‘focuses’ on character features.
• This technique has limited accuracy, it will not copy exact dimples/ freckles/ or t-shirt logos.
• Cref works for both Niji and normal MJ models and can be mixed together. --sref

Advanced features

• You can use multiple URLs to match information/characters from multiple such images. --cref URL1 URL2 (This is similar to multiple image or style prompts)

How does it work on Web Alpha?

• Drag or paste the image into the Imagine bar, it now has three icons. Choosing these sets whether it’s an image cue, a style reference, or a character reference. Select Shift+One option to use an image for multiple categories.

Remember, these and other features may change suddenly while MJ V6 is in alpha, but the V6 official beta is coming soon. We’d love everyone’s thoughts on ideas and features.We hope you enjoy this early release and hope it helps you play with building stories and worlds.