Midjourney founder David Holz discusses the challenges of developing a video model. The company also officially launches a hardware team in San Francisco.
In an interview with Midjourney expert Nick St. Pierre, Midjourney CEO David Holz gives an insight into the development of the company's upcoming video model.
Instead of offering a single high-resolution video, Holz sees the advantage of giving users several lower-resolution videos to choose from and then enhancing the quality. This approach would give users more flexibility and control. In addition, Holz believes that slow rendering can hinder the creative process by depriving creatives of a sense of flow.
Holz notes that Midjourney is still experimenting to find the right balance between speed, cost, and quality. Factors such as reduced resolution or frame rate can create new problems. The company plans to train several versions of the model to find the optimal balance.
For Holz, speed seems to be the key factor. Midjourney has ideas for speeding up the model, which could come at a higher cost, although he acknowledges that a more expensive model would be accessible to fewer people.
"It's hard to say how it'll all balance out," says Holz, who envisions a more expensive, faster flagship model that becomes more affordable over time. Midjourney is also working on a separate 3D model that could later be merged with the video model.
Midjourney launches hardware team in San Francisco
Midjourney also announced the official start of its hardware development. The hardware project will be led by Ahmad Abbas, Head of Hardware at Midjourney, and Midjourney founder David Holz.
Both have extensive experience in hardware development. Holz developed Leap Motion's hand tracking hardware with Abbas, who most recently served as hardware engineering manager for Apple's Vision Pro.
The exact purpose of the project, informally called "Midjourney Orb" and confirmed by Holz earlier this year, is not yet known. It is possible that the hardware will be designed for AI-generated 3D worlds and possibly real-time generated video games.
Holz has previously mentioned that he believes there will be a console in a few years with an AI processor that simulates all games in real time, describing the company's image model as a "really slow game engine" and promising volumetric 3D worlds generated at 60 frames per second in the future.
