Google DeepMind has unveiled ‘Genie’, an AI capable of creating interactive video games from a mere prompt or image. This breakthrough AI draws its experience from 200,000 hours of gameplay videos found on the internet. Impressively, Genie doesn’t require specific action or text annotations to recognize the main character in a game, allowing players to dive directly into the action.
Utilizing its advanced Latent Action Model, Video Tokenizer, and Dynamics Model, Genie represents a significant leap in AI technology. Although currently exclusive to researchers and generating games at a modest 1FPS, the implications are nothing short of revolutionary.
Imagine a future where anyone can conjure up their own virtual worlds at will. Stay tuned for a deeper dive into how Genie works in tomorrow’s newsletter. For now, let the anticipation (and perhaps a few Google memes) begin!
Official Website
The game generated by Genie
robotics video