Midjourney’s Breakthrough: How AI Technology Is Transforming Character Creation

Midjourney has introduced an innovative feature to its AI image generation service, allowing for the creation of consistent characters across multiple images. This long-awaited update answers one of the most frequently requested features by users and addresses a significant challenge that AI image generators have faced up until now​​​​​.

Traditionally, AI image generators that use diffusion models, like the well-known Stable Diffusion algorithm from Stability AI, create new content for each prompt input. This approach made it difficult to maintain consistency of characters across images, posing a particular problem for visual storytelling mediums such as movies, novels, and comics where character continuity is crucial​.

Midjourney’s new feature, tagged “–cref” (short for character reference), enables users to add a URL of a previously generated character image to their text prompts. This guides Midjourney’s algorithm to match the character’s distinct features, body type, and clothing in new images. This feature potentially transforms Midjourney from an ideation tool to a professional tool used by creators​.

Users have some control over how closely the new image resembles the original character by using the “–cw” tag (character weight) with a number from 1 to 100, adjusting the variance of the resulting image. This feature is currently being tested by artists and creators, offering them new possibilities in maintaining narrative continuity and exploring various scenes, settings, facial expressions, and props for their characters​.

Moreover, Midjourney introduces advanced options such as the ability to blend information from multiple images by using several “–cref” tags with their respective URLs. While this feature works best with characters created within Midjourney and is not intended for real people or photos, it does not capture every detail from the original image but focuses on character traits and allows for the combination of multiple character references in a single image​.