A firm claims to be training an AI system that will allow designers to create cinematic environments with total control over the lighting, motion, characters, and scenery. How? by having people hike all over the world while strapping cameras to their backs.
Self-driving pioneers Jeff Hawke and Oliver Cameron founded Odyssey, which claims to have developed a “advanced camera capture system” that can gather data almost anywhere a person can reach. The system, which weighs roughly 25 pounds, has two lidar sensors, six cameras, and an inertial measurement unit.
Similar to Google’s Street View Trekker, the system can record its environment in 360 degrees and at “3.5K resolution,” with metadata containing “physics-accurate” depth information.
What’s the point, then? According to Odyssey, the system’s data is being fed into algorithms in order to “capture the fine details that make up our world.” In essence, the organization is creating digital recreations of real-world landscapes, such as those with parks, buildings, woods, caves, trails, beaches, glaciers, and so forth, a la Meta’s Hyperscape project.
How these reconstructions will result in improved generative tools for creatives is yet unclear. According to earlier statements by Cameron and Hawke, Odyssey has produced a number of generative AI models that build layers of visual information, such as item shape, lighting, and motion, and then merge them into a single virtual “world” to produce the visuals that are required.
However, even the greatest “world models” available today have drawbacks, and Odyssey makes no claims to have resolved them entirely. However, obtaining funds is necessary to move on.
Odyssey said today that EQT Ventures led a $18 million Series A investment round in which GV and Air Street Capital also participated. The additional funding will be used to expand Odyssey’s data collection activities in California, bringing the company’s total fundraising to $27 million.
In the future, Odyssey intends to extend its data collecting to additional states and nations, hopefully with privacy safeguards in place.
“We think it will be impossible for generative models to generate Hollywood-grade worlds that feel alive without training on a vast volume of rich, multimodal real-world 3D data,” the business stated in a blog post.“We believe an advanced generative world-building model will unlock a better way to create film, games, and more.”