Genie 4

Google

2026-10-28 02:05:23

Google DeepMind unveils Genie 4, a real-time world model with audio, dialogue and longer memory

Google DeepMind has introduced Genie 4, its latest general-purpose world model, positioning it as a step beyond video generation and toward real-time world simulation. According to the company, Genie 4 can generate interactive environments from a text prompt, update scenes on the fly, and keep a coherent world running for about two minutes at roughly 24 frames per second. The model also marks the first time the Genie series can natively generate audio and support improvised character dialogue. The release highlights a broader technical shift inside Google’s AI stack. DeepMind says Genie 4 is built as a hybrid world model that works with Gemini 3.8 Flash, Gemini 4 Argon and Gemini Robotics. In the examples shared by Google and early testers, the system handled dynamic scene edits, more realistic human behavior, stronger rigid-body physics, and multi-step robotic tasks, including checkers moves and lab manipulation with a UR5 robotic arm. Genie 4 is now in limited testing, and Google AI Ultra users can apply for the waiting list. Google also said the model will become the first world model available through the Gemini API and Google AI Studio, opening it to developers.

00

Google DeepMi

2026-10-27 17:56:37

Google DeepMind unveils Genie 4 with native audio, dialogue and roughly 2 minutes of coherent interaction

Google DeepMind has introduced Genie 4, a new world model that can generate environmental sound, music and interactive dialogue natively as scenes unfold, while also letting users change weather, camera angles and scene elements through natural-language prompts in real time. The company described it as its most realistic and interactive world model so far and shared a roughly 147-second demo on X in the early hours of Oct. 28 Taipei time. Outside reports from TechCrunch and The Verge said coherent interaction lasts about 2 minutes, roughly double the approximately 1 minute cited for Genie 3. DeepMind itself used more cautious language, saying the model can stay coherent over short spans, while details and object permanence may drift over longer runs. The company said Genie 4 is being released first as a research preview to trusted testers rather than as a broad public launch. DeepMind also tied the system closely to the Gemini family, including Gemini Omni for grounding simulations in real-world appearance and behavior, Gemini 3.8 Flash as a guiding agent in the loop, and experiments involving Gemini Robotics and Gemini 4 Argon. Future access is planned for Project Genie under Google AI Ultra, as well as early partners using the Gemini API and Google AI Studio, with a waitlist also opening.

20