Midjourney's next major update is just around the corner, with the company aiming to release in the upcoming weeks.
During recent office hours meetings in their Discord channel, the company revealed they're rebuilding their system from the ground up, with plans for a new architecture, fresh datasets, and better ways to understand what users are asking for. Right now, they're testing everything internally - a process that could take anywhere from a week to a month. After that, they'll need about two weeks to fine-tune the system.
According to Midjourney, they've essentially started over with nearly every part of their technology. Once they finish fine-tuning, they'll roll out an alpha version with limited features. Looking ahead, they've got several additions planned: different design modes, the ability to process multiple images at once, and specialized versions of v7 that let users choose whether they want faster results or higher quality.
As before, the company continues to explore video generation, working on their own model while considering partnerships with open-source projects. The big challenge, they say, is making video generation enjoyable for users while juggling three key factors: how fast it works, how much it costs, and the quality of the results.
Midjourney is also developing a new approach to 3D generation. Instead of using traditional methods like polygon representation, NeRFs, or Gaussian splatting, they're working on their own "NeRF-like" format. The company hasn't shared when this feature might be available to users.