Odyssey has officially debuted a generative AI model capable of streaming interactive 3D environments in real-time. By leveraging clusters of Nvidia H100 GPUs, the startup is pushing the boundaries of world modeling, allowing users to navigate AI-generated spaces at up to 30 frames per second.
Early Performance and Technical Challenges
While the technology represents a significant leap forward, the company openly acknowledges that its current demo remains in an experimental phase. Users may notice that environments occasionally appear blurry or distorted. Furthermore, the model currently struggles with long-term spatial consistency; as users move forward or rotate, the generated surroundings can shift unpredictably, indicating that the system is still refining its sense of persistent state.
The world played forward, by a model.
On the one hand, it's calm and serene. On the other, it's chaotic and terrifying.
I think the model nailed it in both cases. pic.twitter.com/vuszeL9p0V
— Oliver Cameron (@olivercameron) May 28, 2025
The Road to Stability and Interaction
Odyssey is aggressively optimizing its architecture to address these stability hurdles. The company’s roadmap focuses on developing richer world representations that maintain fidelity over time. Beyond mere visual generation, the research team is expanding the model’s “action space,” moving from simple motion to complex world interactions by training on large-scale video data.
Currently, the cost to run these simulations is estimated between $1 and $2 per user-hour, a figure the company intends to refine as the technology matures.
A Unique Data Strategy
Diverging from industry peers who rely heavily on existing public datasets, Odyssey is utilizing a proprietary approach to data acquisition. The startup has developed a specialized 360-degree, backpack-mounted camera system. By capturing real-world landscapes firsthand, Odyssey aims to build high-quality foundation models that surpass the limitations of models trained strictly on scraped online content.
Investment and Industry Backing
The ambitious project is backed by significant venture capital, with $27 million raised to date from firms including EQT Ventures, GV, and Air Street Capital. The company’s vision is further bolstered by its leadership, which includes Ed Catmull—a co-founder of Pixar and former president of Walt Disney Animation Studios—on its board of directors.
Integration with Creative Workflows
Odyssey is positioning itself as a vital tool for creators rather than just a standalone demo. The company is actively developing software pipelines that allow artists to export scenes generated by its AI into industry-standard tools. This interoperability will enable designers to load, refine, and hand-edit AI-generated environments within platforms like Unreal Engine, Blender, and Adobe After Effects.