Midjourney starts the day by announcing the new option –p for ‘personal custom style’ or something like that.
At first, the excitement is total. Finally, we can customize something! But no. It’s not that nice.
The –p option takes the ranking of images we have made over time, calculates a unique code, and extracts a style by combining the choices we have made when rating images in pairs in the Ranking section of the website.
One moment.
First. What if we have chosen images randomly with the sole purpose of getting free GPU hours? The style that comes out of that is… random.
Second. We have done it ‘right,’ but… we have rated all kinds of images: anime, portraits, landscapes, objects, illustrations, photographs, etc. What do they all have in common? Anything?
We’ll have to see. Midjourney says this ‘model’ evolves slightly as we rate more images. In our case, with more than 243,000 images rated, it will be challenging to turn the style around. It doesn’t matter. At the very least, it doesn’t make sense to mix the rating of a series of realistic photographs with a series of comic book panels. What do they have in common? The aesthetics? No. The theme? Maybe.
In Discord, voices are already rising for this option to be broader. For it to be possible to reset the counter. For it to be possible to choose which images are rated. And a long etcetera.
If there is no certain evolution of that feature, it will remain just another random style in the vast collection of –sref.
At PIxel Editions, we will be keeping an eye on what happens in the coming days.
