Genie 4 generates interactive, audio-enabled simulations steerable by text. Developers can expect Gemini API and AI Studio early access, and agent builders gain a route to simulated training environments.
Key takeaways · AI-distilled
DeepMind runs Gemini 3.8 Flash as an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → in the loop: while Genie 4 renders, Gemini watches incoming frames, splits a goal into steps, checks that each step succeeded, and replans when the user intervenes.
Genie 4 is DeepMind's first real-time world model to generate audio natively, including ambient sound and character dialogue that reacts to user input and world events as the simulation runs.
Users steer a running simulation with text, such as changing the weather or adding a person, and Gemini adapts each prompt to the surrounding scene before Genie renders it.
DeepMind says coherence holds only over short sessions: details and object permanence still drift in longer runs, after many prompt edits, or while moving through the scene, and legible text and spatial accuracy remain active research areas.
Access is limited to trusted testers for now, with Google AI Ultra subscribers in Project Genie and early Gemini API partners next, and all Genie 4 video and audio carries SynthID watermarks.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
