AIO Apex
en

Google Opens Genie 4 World Model to Developers via Gemini API

Google
Share:
Google Opens Genie 4 World Model to Developers via Gemini API

Google opened Genie 4, its real-time interactive world model, to developers on November 19 via the Gemini API and Google AI Studio — one day after the model launched publicly for Google AI Ultra subscribers. The release gives developers API access to the same technology that powers Project Genie, Google's publicly available interactive world generator at labs.google/projectgenie.

Genie 4 generates continuous 720p worlds at 24 frames per second with synchronized native audio, ranging from ambient soundscapes to character dialogue. Developers can steer environments in real time using keyboard controls alongside live text prompts, shifting weather, lighting, characters, or events mid-session without breaking the world's coherence. Google says Gemini reasoning keeps the generated world state consistent throughout a session.

Access is available through the Gemini API's Interactions API and Google AI Studio, using the model ID models/genie-4. Developers can connect via WebRTC from browsers or WebSocket from Python and Node.js. A paid-tier Gemini API key is required; the model is not available on the free tier. Pricing is approximately $0.10 per second of streamed output. Sessions cap at 120 seconds, a constraint tied to the model's rolling window architecture.

Google identifies three primary developer use cases: prototyping interactive experiences, visualizing spatial design concepts, and training robotics and AI agents. The robotics use case is particularly notable — generating diverse simulated environments is one of the most resource-intensive parts of training embodied AI systems, and Genie 4's real-time interactivity could accelerate that process significantly.

The consumer version of Project Genie, launched November 18 for Google AI Ultra subscribers, adds a community gallery where users can publish and remix worlds. That version also brings Genie 3 up to 3-minute extended sessions (up from 60 seconds), 360-degree panorama grounding, and real-time in-world events. Google acknowledges that camera latency, complex physical interactions, real-time audio precision, and long-term scene permanence remain unsolved problems.

Every output — video and audio — includes SynthID watermarking, and inputs and outputs pass through safety screening. Prompts involving real people, public figures, or brand IP can be blocked.

An interactive canvas and starter apps are available in Google AI Studio at aistudio.google.com/live?model=genie-4. For complete API documentation, see aistudio.google.com/docs/genie.

As announced by Google on November 19, 2026 on the Google for Developers blog.

Originally reported by Google. Read the original article for additional details.

View original source
Share: