Midjourney just made it easier to generate AI images the way you want them with a new ‘personalize’ feature. Using it overrides the style biases the AI model developed in training. In most cases, there won’t be much difference, especially if you’re not a particularly heavy user of Midjourney — but those putting in the time to generate images can make something unique.
Before you can use Personalize you need to work through an image ranking process, where you compare two Midjourney-generated pictures and pick your favorite. This trains your personalization algorithm and feeds into the image-generation process. Once you’ve trained up your personalization model you can then add --personalize to the end of your prompt and it will generate based on the preferences you selected while ranking.
To put Personalize to the test I came up with five varied prompts and ran each of them twice, once with and once without the setting. What it showed is I think I need to spend more time ranking images to further fine-tune my algorithm. If you are using Midjourney web it is just a switch available under the create settings at the time of writing your prompt.
If you use Discord you need to add --personalize to the end. This will also generate a code you can share with others so they can try out your own personal style. First up I wanted to see if I could get it to create a smartphone-style photograph of something most people wouldn't be able to resist photographing — a beautiful ma.