Midjourney now allows users to personalize image generation to their individual preferences, considering user ratings and likes.
Midjourney has released an initial test version of personalized image generation. The new feature aims to more closely align the generated images with each user's unique tastes.
According to Midjourney, many aspects of each prompt remain undefined. Until now, Midjourney's algorithms have filled these gaps with their own "preferences", which generally reflect the combined preferences and biases of the Midjourney community.
But every user has their own personal tastes. This is where model personalization comes in: It learns what the individual user likes and is more likely to fill in the prompt gaps based on their preferences.
Currently, personalization is based on the ratings users provide when comparing images in pairs and the images they "like" on the Explore page. At least 200 such ratings or likes are required to enable the feature.
Users can check the number of their own ratings by using the "/info" command or visiting the personalization page.
To use personalization, you must add "--p" after the prompt or enable personalization for all prompts in the settings under "/settings" or via the "Prompt settings" button on the website.
When personalization is enabled, a "code" is appended to the prompt. This code can be shared with others to pass on the personalization effect applied to the image.
The strength of the personalization effect can be adjusted using the "--s" parameter, where 0 corresponds to "off," 100 to "standard," and 1000 to "maximum."
Midjourney emphasizes that personalization is not yet a stable feature and will change (subtly) with additional user ratings. Since algorithm updates may be rolled out in the coming weeks, users should consider personalization a "fun feature constantly in flux."
