‘Style Reference’ (SREF) is one of the cutting-edge features launched by Midjourney. SREF operates like a style learning LoRA, permitting you to persistently generate images in a particular style simply by learning from an image supplied. Just one image is needed to achieve consistent results.
To utilize this remarkable feature, start with a prompt and after that, apply the tag ‘–sref’, subsequently followed by the URL of an image; more than one image can be used. Midjourney will then assess the image(s) provided and generate your image reflecting the style and aesthetics of the image supplied. This feature functions well with both V6 and Niji V6 models, but does not support earlier models.
Advanced settings of SREF allow users to use multiple images by using the argument ‘–sref’ and separating images with a space. There is also an option to set weights to styles by adding a double colon and the weight after the image URL. You can manage the overall power of the stylization by the ‘–sw’ argument; 100 is the default setting, 0 switches it off and 1000 sets it to maximum. In the case of using image prompts, your images should precede ‘–sref’.
Midjourney essentially grades the color of your generated image using the color scheme of your reference image, hence allowing you to imitate the aesthetics of a reference image. For instance, by using a reference image from Mad Max: Fury Road, you can create an image of a modified 1965 post-apocalyptic mustang in a dystopian scene with the same color grading as the reference image, simply by using the given prompt in the user’s guide.
This proves beneficial in achieving a particular aesthetics and style as Midjourney interpretations can vary when not using a ‘–sref’ style reference. As the examples given in the article from the movie ‘Tenet’ show, SREF style reference enables you to emulate the aesthetics of a reference image, basically replicating the color grading of the reference image. This feature affords consistency in appearance and aesthetics which can be extremely advantageous, especially for series of images where maintaining a particular style is essential.
This innovation is essentially what sets Midjourney apart, as it allows effective replication of colour grading from reference images for consistently styled images. A bunch of samples are available for users to explore.
