Odyssey-2
Odyssey-2 is a general purpose world model that generates real time interactive video simulations from text or image prompts. It predicts how physical environments evolve through continuous, action conditioned rollouts, producing over two minutes of coherent video while responding to user input in real time. The model learns physics implicitly through next state prediction rather than an explicit physics engine, using a causal autoregressive diffusion transformer architecture.
- Type
- Video
- Weights
- Open
More from Odyssey
6
-
Agora 1
Video
Agora-1 is Odyssey’s multi-agent world model that lets multiple human or AI participants share and interact i…
-
Odyssey 3
Multimodal
A foundation world model trained on visual observations of the world. It learns physics, dynamics, cause and…
-
Starchild 1
Multimodal
Starchild-1 is Odyssey’s real-time multimodal world model that generates synchronized audio and video while r…
About Odyssey
Odyssey is a US AI research lab building general-purpose world models — AI systems that generate real-time, interactive simulations of arbitrary environments. Products include Odyssey-2 (general-purpose world model), Starchild-1 (multimodal world model), Agora-1 (multi-agent shared simulation), and PROWL (RL adversarial framework). Applications span gaming, robotics, defense, education, and film. API-accessible. Raised $27M+ from EQT, GV, NVIDIA, and Samsung Next.