A London-based AI research lab, Odyssey, has launched a groundbreaking research preview that transforms traditional video into fully interactive, real-time virtual experiences. This early prototype, while still experimental, may represent a completely new entertainment medium—something Odyssey dubs “an early version of the Holodeck.”

With initial applications in film production and game development, Odyssey’s technology opens the door to AI-generated interactive videos that respond instantly to your actions—whether via keyboard, controller, smartphone, or even voice.

What Is AI-Generated Interactive Video?

Unlike traditional video games or CGI, Odyssey’s tech is powered by an advanced AI world model that creates video content dynamically, frame by frame. Rather than relying on pre-programmed responses, this system predicts the next video frame based on user input and the current digital environment—similar to how a language model predicts the next word in a sentence, but far more complex.

According to Odyssey, their AI generates realistic-looking video frames every 40 milliseconds, enabling almost instant interaction and a highly immersive user experience.

“The experience today feels like exploring a glitchy dream—raw, unstable, but undeniably new,” says the Odyssey team.

The Tech Behind It: What Is a World Model?

At the core of Odyssey’s innovation is something called a world model, technically defined as an action-conditioned dynamics model. Here’s how it works:

• The model observes the current video state.
• It receives the user’s input (a key press, click, or voice command).
• Using its training data and understanding of past events, it predicts and generates the next frame.

This real-time prediction loop makes each interaction organic and unpredictable, offering a far more fluid experience than traditional, logic-based games where every action is scripted.

Tackling Stability: Solving AI Drift with Smart Training

One of the biggest challenges in AI-generated video is “drift”, where minor errors in frame prediction build up over time, degrading the quality or stability of the video.

Odyssey addresses this with a “narrow distribution model” approach:

• First, the AI is trained on broad, general video datasets.
• Then, it is fine-tuned on specific environments for greater reliability.

The result? Improved stability and coherence, though at the cost of some variety. Odyssey says they are already progressing toward a next-gen model with more realistic textures, smoother dynamics, and richer interaction.

Running on H100 GPUs: Real-Time Performance at Low Cost

Running this interactive AI video system in real time is no small feat. It currently costs about £0.80 to £1.60 per user-hour (~$1–$2), powered by distributed clusters of H100 GPUs across Europe and the U.S.

Compared to the massive costs of producing AAA games or cinematic content, this approach could soon become a cost-effective alternative. Odyssey also predicts these operational costs will drop significantly as the model becomes more efficient.

Is This the Future of Storytelling?

Every major technological leap—books, radio, film, video games—has transformed how we tell stories. Odyssey believes AI-generated interactive video is the next frontier.

This new medium could reshape:

• Entertainment: Interactive movies and immersive narratives.
• Education: Skill-training videos that respond to learner input.
• Tourism: Explore destinations virtually, with real-time control.
• Advertising: Personalized, dynamic product demos.

Though still in its early stages, Odyssey’s prototype offers a compelling glimpse into the future—where users won’t just watch stories, but live them.

Final Thoughts: Why This Matters

Odyssey’s real-time AI video model could redefine not just how we consume content, but how we interact with digital environments. It bridges the gap between video games, movies, and immersive simulations—creating something entirely new.

As AI continues to evolve, technologies like these may become standard across streaming platforms, education systems, and even virtual social experiences.

Keep an eye on Odyssey—they might just be building the next big thing in AI entertainment technology.