Midjourey, the popular generative artificial intelligence (AI)-powered image generation tool, has announced a new feature that will allow users to take a reference character and use it across multiple different images. The feature, called Character Reference, is currently in testing and only available on the Midjourney 6 (MJ6) and Niji 6 models. The feature is similar to the AI tool’s existing Style Reference command but instead of the style of the image, it keeps the character of the image consistent.
The announcement was made by the official account of Midjourney on X (formerly known as Twitter), where it highlighted that the under-testing feature is only available to the MJ6 and Niji (for anime and illustrative styles) models. The feature can now be tested by users on Discord, where Midjourney officially operates. Interested people will need a subscription to the platform in order to generate AI images.
Explaining the Character Reference feature, Midjourney highlighted that it can be used in prompts using the cref command. The command should be followed with a URL to an image of a character. Characters designed using Midjourney will offer the best results. The company also mentions that the feature will not work for real people or photos and the AI is likely to distort them. Further, the precision of the feature is currently limited and it will not copy minor details such as dimples, freckles, or t-shirt logos.
For more control over the character, users can use the cw command for Character Weight. Adding a value between 0 to 100 will increase the consistency of the character. While at 0, the focus will just be on the face, 100 will keep the face, hair, and clothes consistent. Notably, Midjourney is a very prompt-sensitive model, and with the introduction of MJ6, the importance of grammar and including the right prompts has increased.
Everything about an image can be controlled using specific commands and the language of the prompt. So, to ensure character consistency, the cw command should be used, and the prompt should not be kept vague. Some of the use cases of this feature could be for illustrations in a story, creating a storyboard for digital comics, or creating videos with background music and audio.
