A startup is training an AI model that it claims will enable creators to generate cinematic worlds, with complete control over the scenery, characters, lighting, and motion. How? By having humans strap cameras to their backs and hike around the world.
Odyssey, founded by self-driving pioneers Oliver Cameron and Jeff Hawke (Cameron was previously the VP of product at Cruise), says it’s created an “advanced camera capture system” that can collect data just about anywhere a person can reach. Weighing about 25 pounds, the system packs six cameras, two lidar sensors, and an inertial measurement unit.
Bearing a resemblance to Google’s Street View Trekker, the system can capture its surroundings in “3.5K resolution” and 360 degrees, with “physics-accurate” depth information metadata attached.
So what’s the point? Well, Odyssey says it’s taking data from the system and feeding it through algorithms to “capture the fine details that make up our world.” Essentially, the company’s generating digital reconstructions of real-world scenes — scenes with forests, caves, trails, beaches, glaciers, parks, buildings, and so on.
Now, it’s not totally clear how these reconstructions will translate to better generative tools for creatives. Cameron and Hawke have previously said that Odyssey has developed several generative AI models that create layers of visual detail, including object geometry, lighting, and motion, and then combine these into a single virtual “world” to create desired scenes.
Even the best “world models” today have limitations, however — and Odyssey doesn’t claim to have solved all these. Nevertheless, it’s securing cash to forge ahead.
Odyssey today announced that it raised $18 million in a Series A funding round led by EQT Ventures with participation from GV and Air Street Capital. The new money, which brings the company’s total raised to $27 million, will be put toward scaling up Odyssey’s data collection operations in California.
Odyssey plans to expand its data collection to other states and countries in the future — with privacy protections in place, one would hope. (Google’s Street View team, for one, has found itself in the crosshairs of regulators for capturing images of public places that violated people’s privacy.)
“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 company wrote in a post on its blog. “We believe an advanced generative world-building model will unlock a better way to create film, games, and more.”
