On November 19, Google announced that it is officially opening its latest world model models/genie-4) to paid developers through the

Genie4 is built on Gemini Omni's visual rendering and the reasoning capabilities of the Gemini series. It supports generating continuous HD interactive worlds at 720p, 24fps from text or reference images, and simultaneously outputs native stereo audio including ambient sound, sound effects, and character dialogue.
Users can move in real time within the scene via WASD or arrow keys (response latency of about 0.5 seconds), and dynamically adjust weather, lighting, and events with real-time text prompts (latency of about 1 to 2 seconds), while the system uses its reasoning mechanism to maintain the coherence of the world state. In addition, AI Studio has simultaneously launched an interactive canvas and a starter app to lower the development barrier, and all generated content is embedded with an imperceptible SynthID watermark.
In terms of technical architecture and pricing, Genie4 uses single-session temporary tokens to ensure key security. The browser and server sides use WebRTC and WebSocket respectively for streaming transmission. A single session is capped at 120 seconds, and service covers data centers in the United States, Western Europe, and Japan. Its API pricing is $1.50 per million input tokens, $9 for text output, and $17.50 for video output, with a combined streaming cost of approximately $0.10 per second.
As a commercialization continuation of Google DeepMind's October research preview, Genie4 pushes world simulation capabilities from internal research into the third-party application ecosystem, not only providing new tools for interactive prototyping and spatial visualization, but also offering a low-cost, high-fidelity safe training environment for robots and artificial general intelligence (AGI) agents.
Join Now