Google DeepMind Launches Genie 4 With Native Audio and Real-Time World Editing
The experimental world model generates interactive environments with sound and responsive dialogue, with access initially limited to trusted testers.
Published
Last updated / Fact-checked on
Google DeepMind announced Genie 4 on Oct. 27, introducing a real-time interactive world model that generates explorable environments alongside music, ambient sound and dialogue. The model is launching as a research preview for trusted testers, with broader access planned through Project Genie, the Gemini API and Google AI Studio.
The release expands DeepMind's work on simulations that respond to user actions rather than simply play back a generated sequence. Users can navigate with keyboard controls and modify an unfolding scene through natural-language instructions, changing the weather, moving the camera or introducing people and objects. DeepMind says Gemini models help interpret those requests and maintain continuity as the environment changes.
Worlds That Respond and Speak
Native audio is a central addition. DeepMind describes Genie 4 as its first real-time world model to generate sound directly, including environmental noise and character dialogue that adapts to user inputs and events. Its published examples span realistic settings and stylized environments, including a landscape made entirely of stained glass.
The company also claims improvements in physical interactions and human movement, including simulations involving aerodynamics and collisions. Characters are designed to move more naturally, respond to their surroundings and accept real-time direction. These remain company-described capabilities rather than proof that the model reliably reproduces real-world physics.
TechCrunch and The Verge reported that Genie 4 can sustain a consistent scene for about two minutes, double the roughly one minute of its predecessor, Genie 3. DeepMind's announcement, however, emphasizes coherence over short sessions and acknowledges that longer interactions can produce inconsistencies.
Gemini Helps Guide the Simulation
Genie 4 draws on several models in Google's Gemini family. According to DeepMind, Gemini Omni helps anchor simulations in real-world appearance and behavior, while Gemini 3.8 Flash acts as a guiding agent: observing generated scenes, breaking goals into steps and checking incoming frames to assess progress.
That arrangement is intended to support events requiring more than a single action. As users intervene or conditions change, the guiding model can adjust its plan while Genie continues generating the environment. DeepMind also described experiments combining Genie 4 with Gemini 4 Argon and Gemini Robotics, including a simulation of Apptronik's Apollo humanoid playing checkers.
The company presents this combination as an early step toward hybrid world models that pair structured reasoning with generated visual and audio environments. Potential applications include training scenarios for robots and AI agents, although the announcement does not establish that these experiments translate into reliable real-world performance.
Creative Uses, Experimental Limits
DeepMind says trusted testers are exploring filmmaking, journalism, education, gaming and engineering workflows. Examples include adjusting camera angles and dialogue during scene previsualization, turning archival photographs into explorable environments and prototyping game concepts before writing engine code. Enterprise partners are also evaluating rapid scenario generation.
Those possibilities come with substantial limitations. Visual details and object permanence can drift, particularly during longer sessions, complex edits or movement through a scene. DeepMind identifies complex physical interactions, audio accuracy, readable text, spatial accuracy and long-term consistency as ongoing research challenges.
The company says it conducted internal safety evaluations before the preview and will embed its SynthID watermark in generated video and audio to help identify AI-created content.
What Comes Next
For now, the launch is a restricted research preview, not a general release. DeepMind says Genie 4 is coming soon to Google AI Ultra subscribers through Project Genie and to early-access partners using the Gemini API and Google AI Studio. A Project Genie waitlist is open.
The immediate test is whether testers can turn more responsive simulations into useful workflows despite the acknowledged consistency problems. DeepMind is positioning the preview as a way to gather that evidence before expanding access.
Sources
- Google DeepMind, "Genie 4: A new generation of world simulation," Oct. 27, 2026: https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
- Google, "We're announcing Genie 4, our new world model for real-time interactive simulation," Oct. 27, 2026: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-genie-4/
- TechCrunch, "Google's Genie 4 world model adds audio, more realistic worlds, and real-time steering," Oct. 27, 2026: https://techcrunch.com/2026/10/27/googles-genie-4-world-model-adds-audio-more-realistic-worlds-and-real-time-steering/
- The Verge, "Google's next Genie AI model generates more realistic experiences," Oct. 27, 2026: https://www.theverge.com/tech/1019076/google-genie-4-ai-model