Google DeepMind releases Genie 4 world model with real-time interactive 3D environments

Oct 27, 2026

Google DeepMind released Genie 4 on October 27, 2026, advancing its AI world model technology to generate photorealistic 3D virtual environments with integrated real-time audio and interactive prompt edits. The model uses Gemini 3.8 Flash as a guiding agent to plan actions and keep simulations coherent for up to two minutes. Released as a research preview, Genie 4 targets applications in game development, film prototyping, and robotics training amid broader consolidation across the AI world model industry.

Release details and availability

▪Google opened a waitlist for users to apply to try the Genie 4 model in Project Genie
▪Google DeepMind released its Genie 4 AI world model in research preview for trusted testers, with plans to roll it out to Google AI Ultra subscribers in Project Genie and early access API partners

Interactive audio and visual capabilities

▪Google DeepMind's Genie 4 allows users to alter simulated scenes in real time by changing weather conditions, modifying camera angles, or adding new people and objects
▪Google DeepMind's Genie 4 generates real-time audio that matches simulated environments, including ambient noises, music, and interactive dialogue between characters
▪Google DeepMind's Genie 4 generates photorealistic 3D virtual environments that users can explore using keyboard controls and alter mid-simulation using text prompts
▪Genie 4 is Google DeepMind's first real-time world model capable of generating audio and images together dynamically in response to user inputs

Performance improvements over Genie 3

▪Google DeepMind's Genie 4 maintains consistent virtual scenes for up to two minutes, doubling the one-minute limit of Genie 3
▪Google DeepMind reported that Genie 4 provides improved physics simulation and renders more human-like characters than Genie 3

Integration with Gemini models

▪Google DeepMind tested its Gemini 4 Argon model alongside Genie 4 to simulate an Apptronik Apollo humanoid robot playing checkers without using physical hardware
▪Genie 4 builds on capabilities from Google DeepMind's Gemini Omni video model to render realistic scenes, while Gemini models handle background planning to maintain world coherence
▪Google DeepMind experimented with Gemini 4 Argon for Genie 4 simulations, but selected Gemini 3.8 Flash to power Genie 4's real-time experience
▪Google DeepMind integrated Gemini 3.8 Flash into Genie 4 as a guiding agent that evaluates incoming frames, breaks down goals into steps, and keeps simulations on track

Target applications and industry use cases

▪Google DeepMind pitched Genie 4 commercial applications across interactive gaming, film prototyping and storyboarding, interactive learning, and physical robot training environments
▪Game developers and filmmakers are using Genie 4 to turn creative concepts into explorable prototypes and sketch out scenes

Industry landscape and competitor acquisitions

▪AI video startup Runway released its second playable world model, GWM Worlds 2, featuring dynamic user interaction and audio generation prior to the release of Genie 4
▪AI startup World Labs, founded by Fei-Fei Li, is being acquired by chipmaker AMD as part of broader AI world model industry consolidation
▪Amazon entered talks to acquire AI company Decart, while electric vehicle manufacturer Rivian acquired Third Dimension to integrate simulation capabilities for self-driving testing