Content
summary Summary

Midjourney released an early version of its new AI model customization system that lets users create multiple profiles and customize their AI models more quickly using collections of reference images.

Ad

The new infrastructure marks a significant upgrade to the personalization features the company introduced in June. It needs far fewer image ratings to become effective. Users can start with as few as 40 ratings, and profiles become stable at around 200 ratings.

While 2,000 ratings still provide better results, Midjourney founder David Holz says "you shouldn't practically need 10,000 anymore."

The system also includes a new mood board feature that lets users upload multiple reference images to guide the AI's output. Holz explained that diverse image collections lead to more complex combinations in the generated results.

Ad
Ad
Mood board with two columns: on the left, space scenes, planets and space stations; on the right, nature shots with Yosemite rocks, falcons and stylized Christmas trees.
Image: Midjourney

Users can add images to mood boards by uploading them directly, adding URL links, or selecting from their library of previously generated images. The mood boards can be shared with other users through parameters.

The update also brings support for multiple model profiles, allowing users to name and set default profiles for different projects. Users can now also view all images created with a specific profile, making it easier to track outputs across different customizations.

Mood board with 1970s vacation photos: canoe, VW Beetle, mountains, lakes, nature scenes, European architecture and Christmas decorations.
Image: Midjourney

Users can access the new personalization features through midjourney.com/personalize once they've completed at least 40 image ratings.

Year-end AI competition intensifies

The release comes amid a flurry of year-end AI announcements from major AI labs. OpenAI launched its "12 Days of OpenAI" campaign featuring various updates, including the Sora video model. Google also recently updated its Veo 2 video model and Imagen image generator. Veo 2's video model outperformed Sora in human ratings.

Midjourney itself recently introduced Patchwork, a collaborative AI-powered canvas tool for creating fictional worlds, currently available as a research preview. The company plans to release its v7 image model and a new video model after the Christmas holiday.

Ad
Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Recommendation
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Summary
  • Midjourney has introduced an early version of its new model personalization infrastructure, allowing users to create multiple profiles, customize models more quickly, and upload mood boards with images for inspiration.
  • According to Midjourney founder David Holz, the personalization profiles require fewer reviews to get started, with 200 ratings resulting in a fairly stable profile, although 2,000 ratings provide even better results.
  • Mood boards, inspired by uploaded or collected images, serve as an alternative to style references in prompts. The more diverse the images, the more complex the models remix them. Users can initiate personalization through the Midjourney website.
Jonathan works as a freelance tech journalist for THE DECODER, focusing on AI tools and how GenAI can be used in everyday work.
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.