MidJourney, an artificial intelligence image generation platform, is working to develop new 3D, video, and real-time creation models that can simulate the entire world.
The details emerged as part of Discord’s recent “Office Hours,” where company leaders answered questions from users about the future of the platform.
Rumors have been circulating for a while that MidJourney is expanding beyond images, but this seems to suggest that the new movement is firmly focused on creating world models.
As they built new models, they added and upgraded other features such as image descriptions and consistent text.
The exact nature of this is unclear, but the company has hinted that it will be an “open world sandbox”, a virtual environment where people can build video games and shoot movies.
Beyond images with AI
MidJourney was founded in 2022 by David Holz, a self-proclaimed serial entrepreneur.
The first open beta was released in July of that year, followed by rapid iterations and upgrades over the next year, with the most powerful v6 model released last December. During the same office hours, Holtz said v7 would be a bigger leap than 5 to 6.
MidJourney remains stubbornly within Discord as a way to manipulate and create images using its model, but some high-intensity users can use the web interface.
Work has already begun on training the seventh version of the MidJourney image model, and Holtz told users on Discord that 3D generation could come before video.
The value of moving to a 3D world
To be clear, we are not currently working on headsets (we are working on hardware related to data capture), but we built an AR headset at Leap Motion over 6 years ago and are open to I made it the source (image and URL are below). : https://t.co/8mX5QTddL9 Open source website:… pic.twitter.com/Z7Bz2WMy0JMarch 13, 2024
It makes some sense that dimensionality is part of the focus. One of Holtz’s former startups was his Leap Motion, a company that built sensors that track hand and finger movements in VR.
“We’re serious about world simulation. We’re building 3D mid-journeys, video MJ, and real-time MJ where things move really fast,” he said on Discord.
When you put all this together, you get a simulation of the world, he added. The goal, he said, is to build on each of these three elements independently and work on how to integrate them.
“It will be more of a sandbox,” he explained. Additionally, he added, “People will create video games and shoot movies in it, but the goal is to build an open-world sandbox.”
On the surface, this sounds like a modern interpretation of the Metaverse. The idea of a fully interconnected set of virtual environments where people can perform daily tasks, earn money, and have fun.
But this seems like a new way to create any type of content. This includes generating an entire virtual environment in which his future science fiction films could be shot without having to use VFX tools to create all the elements at huge cost.
Holtz’s sandbox vision for MidJourney is very similar to some of the ideas Roblox is proposing for development studios.
The company recently announced a new texture and avatar generator. In a conversation with Stefano Corazza, the studio head heard that there are plans for AI to generate entire environments from text prompts.
