The ability to maintain consistent characters or a consistent style from image to image is one of the main challenges for users of generative image AI. The creators of these AIs are working hard to develop solutions.
Since the manual was published, important new features have been released.
1) Style Tuner (Midjourney)
Style Tuners allow you to build a "style seed" from a few generated images. This seed can then be used in prompts via the "–style" parameter to guide generation and achieve a similar style.
To learn more about Style Tuners, check out the official Midjourney documentation.
2) Style Reference (Midjourney and Leonardo)
Style Reference allows you to integrate one or more reference images into the prompt, from which the style will be extracted and replicated in the generated image. Unlike classic image-to-image techniques that integrate an image directly into the prompt to serve as overall inspiration (style, posture, composition...), Style Reference integrates an image into the prompt to serve as style inspiration only. The result is much more precise than with classic image-to-image.
On Midjourney, Style Reference works via two new parameters: –sref and –sw
On Leonardo, it can be activated from the "Image Guidance" tab, in the premium version only.
3) Character Reference (Midjourney)
Character Reference allows you to integrate a reference image containing a character into the prompt, which will be reused in the generated image. This allows you to maintain the same character across multiple images!
The Character Reference works via two new parameters: –cref and –cw
Character Reference and Style Reference can be combined.