AI News · Models ·

Google DeepMind previews Genie 4 world model with audio and live edits

Google DeepMind previews Genie 4 world model with audio and live edits

Google DeepMind released Genie 4 as a research preview, calling it its most photorealistic and interactive world model so far. The model generates matching real-time audio, accepts prompts mid-simulation, and keeps scenes consistent for up to two minutes, double Genie 3. It will roll out to Google AI Ultra subscribers and early Gemini API partners in the coming weeks.

Key points

  • Genie 4 adds real-time audio and lets users edit scenes with prompts mid-simulation.
  • Scenes stay consistent for up to two minutes, double Genie 3's limit.
  • Rollout to AI Ultra subscribers and Gemini API early access partners comes in weeks.

What happened: Google DeepMind released Genie 4 as a research preview for trusted testers, TechCrunch reported. The company describes it as its most photorealistic and interactive world model yet. Users can type prompts while a simulation runs to change the weather, camera angle, or add people and objects.

The details: Genie 4 generates real-time audio that matches the simulated world, including music, ambient noise and dialogue. Scenes stay consistent for up to two minutes, double the limit of Genie 3, which launched a little over a year ago. Google also highlighted better physics and more humanlike characters. In a research experiment linked to Gemini Robotics, a Gemini model planned task steps that Genie 4 simulated in real time, pointing to robot testing before real world deployment. Research director Tejas Kulkarni said the company expects more hybrid world models combining Gemini with generated environments.

Who it affects: Google pitched use cases in gaming, interactive learning and robotics, and testers see potential for film prototyping and storyboarding. In the coming weeks the model will reach Google AI Ultra subscribers through Project Genie, plus early access partners of the Gemini API and Google AI Studio. Google has not disclosed pricing.

What to watch: DeepMind engineer Diego Rivas said it is early for Genie 4 and the end to end product experience is unclear. The team acknowledged that details can drift in longer simulations and that audio accuracy and legible in-world text are unsolved. Competition is growing, with Runway shipping GWM Worlds 2 and AMD acquiring World Labs.

Our take

Teams in gaming, training and robotics now have a clearer path to test interactive simulations, though Google says the product experience is still early.

Sources