Odyssey, a generative AI startup founded by self-driving pioneers Oliver Cameron and Jeff Hawke, has unveiled a groundbreaking AI model capable of streaming interactive video and 3D worlds in real-time. This innovative technology allows users to explore AI-generated environments, blurring the lines between traditional video and interactive experiences, and signalling a significant leap in generative AI capabilities.
Odyssey's Interactive Video Technology
Odyssey's new AI model introduces a novel form of interactive video, where AI-generated sequences respond instantly to user input. Unlike conventional video models that generate an entire clip at once, Odyssey's "world model" updates frame by frame, reacting to user actions such as keyboard or controller input. This allows for dynamic, explorable environments akin to 3D-rendered video games.
Key features of Odyssey's interactive video:
Real-time Interaction: Streams video frames every 40 milliseconds, allowing for immediate response to user input.
World Model Approach: Predicts the next state of the world based on current scene, user actions, and historical data, creating dynamic and interactive environments.
Spatial Consistency: Aims to generate pixels that feel realistic and maintain spatial consistency, even with user exploration.
No Game Engine: Operates without a traditional game engine, relying on sophisticated neural networks.
The Vision for Interactive Media
Odyssey envisions a future where interactive video transforms various sectors, including entertainment, advertising, education, and training. The company believes that its technology can enable entirely new forms of entertainment where stories can be generated and explored on demand, free from the constraints and costs of traditional production.
Future Applications: Potential uses include interactive games, movies, and realistic simulation environments for training robots.
Collaboration, Not Replacement: Odyssey has pledged to collaborate with creative professionals, aiming to provide new tools rather than replace human talent. This commitment is underscored by the involvement of Ed Catmull, co-founder of Pixar, on Odyssey's board of directors.
Technical Underpinnings and Challenges
Odyssey's model is trained on real-world landscapes captured by the company's custom-designed, 360-degree, backpack-mounted camera system. This unique approach is intended to produce higher-quality models compared to those trained solely on publicly available data.
While the demo is still in its early stages and has some limitations, such as occasional blurriness, distortion, and temporal instability, Odyssey is committed to rapid improvements. The system currently streams video at up to 30 frames per second, powered by Nvidia H100 GPU clusters, with costs ranging from $1 to $2 per user-hour.
Key Takeaways
Odyssey's interactive AI-generated video and 3D worlds represent a significant advancement in generative AI.
The technology allows for real-time interaction and exploration of dynamic environments.
Odyssey aims to revolutionise various industries by enabling on-demand, interactive content creation.
Despite current limitations, the company is focused on enhancing visual fidelity, spatial consistency, and user interaction.
The startup has secured $27 million in funding and boasts industry veterans on its board, signalling strong confidence in its potential.
