started · updated
Google DeepMind unveils Genie 4 interactive world model
Google DeepMind has released a research preview of Genie 4, its latest generative world model capable of creating interactive, navigable virtual environments from text prompts. Unlike traditional video generators that produce static clips, Genie 4 allows users to interact with and modify simulations in real time using keyboard controls and natural language commands. Users can change weather conditions, adjust camera angles, or add new characters and objects while a scene is running.
A major advancement in this version is its native multimodality; the model generates synchronized ambient noise, music, and interactive dialogue alongside photorealistic visuals. The model can maintain environmental consistency for up to two minutes, doubling the capacity of its predecessor, Genie 3.
The system utilizes the Gemini model family to function as a director, where Gemini plans steps and checks frames to ensure coherence while Genie renders the visuals. This integration has already been used in experiments with Gemini Robotics to simulate complex tasks, providing a safe virtual environment for testing robot behavior.
While promising for gaming, filmmaking, and robotics, Google noted technical challenges, including potential object drift and loss of detail during extended simulations. Genie 4 will eventually be made available to Google AI Ultra subscribers and early access partners via the Gemini API and Google AI Studio.
Entities
Alphabet · Diego Rivas · Gemini · Gemini 3.8 Flash · Gemini 4 Argon · Genie 4 · Genie 4 · Google · Google DeepMind · SynthID · Tejas Kulkarni
Claims
What the coverage asserts, and how many sources carry each claim.
- [● 35 SOURCES] Google DeepMind released Genie 4 as a research preview. www.whatsnewonthenet.com · www.tokenpost.kr · news.mydrivers.com · ai.cnmo.com · www.bloomberght.com · +28 more
- [● 27 SOURCES] Genie 4 is being tested for use in robotics training. www.whatsnewonthenet.com · news.mydrivers.com · ai.cnmo.com · www.bloomberght.com · www.ithome.com · +20 more
- [● 11 SOURCES] Genie 4 uses SynthID to watermark generated audio and video. www.tokenpost.kr · news.mydrivers.com · swedroid.se · thebridge.jp · www.newsbomb.gr · +6 more
- [● 13 SOURCES] The model maintains a consistent environment for up to two minutes. www.whatsnewonthenet.com · news.mydrivers.com · ai.cnmo.com · www.ifanr.com · www.forux.it · +7 more
- [● 16 SOURCES] The model simulates physical phenomena like aerodynamics and collisions. jornal24horas.com.br · www.takvim.com.tr · www.ifanr.com · www.techeblog.com · www.tokenpost.kr · +10 more
- [● 33 SOURCES] Genie 4 generates ambient noise, music, and dialogue alongside visuals. www.whatsnewonthenet.com · www.tokenpost.kr · news.mydrivers.com · www.iscihaber.net · www.takvim.com.tr · +26 more
- [● 13 SOURCES] The model can experience object drift and loss of detail during long simulations. www.tokenpost.kr · news.mydrivers.com · www.ithome.com · thenewstack.io · donanimgunlugu.com · +7 more
- [● 24 SOURCES] Gemini models act as a director to plan steps and check frames. www.whatsnewonthenet.com · news.mydrivers.com · ai.cnmo.com · www.bloomberght.com · www.ithome.com · +18 more
- [● 18 SOURCES] Users can change the weather or camera angle mid-simulation using text prompts. www.whatsnewonthenet.com · www.moguravr.com · www.technaujienos.lt · www.tokenpost.kr · www.iscihaber.net · +13 more
- [● 3 SOURCES] All generated audio and video content is embedded with SynthID watermarks. www.moguravr.com · www.tokenpost.kr · news.mydrivers.com