ai/list101

Genie 4

deepmind.google

Create and steer interactive simulated worlds in real time with natural-language instructions and native audio.

World Modelsgames
Visit website
Screenshot of Genie 4 official website
Pricing
Not confirmed
Available on
Hosted research preview
Category
World Models

What is Genie 4?

Genie 4 is Google DeepMind's world model for real-time interactive simulation. Users can steer evolving scenes with natural-language instructions, changing weather, moving the camera, or introducing characters. Worlds generate native audio alongside their visuals, including ambient sounds, music, and interactive dialogue.

Access is currently limited to trusted testers in a research preview. Google has announced future access through Project Genie for Google AI Ultra subscribers and early access partners using Gemini API and Google AI Studio. Creative exploration and research into environments for robots and AI agents are proposed applications.

Key features

Live scene direction

Use text instructions during a simulation to change the environment, introduce characters, direct events, and adjust camera movement.

Native world audio

Generate ambient sound, music, and interactive dialogue as the simulated world evolves.

Interactive navigation

Official demonstrations use standard keyboard navigation and camera controls, alongside text prompts that steer the changing scene.

Gemini-supported simulation

Gemini models help interpret and reason about events, supporting the simulated world's appearance, behavior, coherence, and continuity.

Watermarked generated media

Generated video and audio include Google's SynthID watermarking, and the preview applies safety guardrails.

Who is it for?

Creative scene exploration

Creative teams with preview access can explore how an environment changes when they alter its weather, characters, or events.

Interactive experience concepts

Game designers with preview access can investigate scene ideas and navigation concepts before committing to a production implementation.

Simulation research

Researchers with preview access can assess whether generated environments suit exploratory work involving AI agents or robotics.

How to use Genie 4

  1. Review the official product page and demonstrations to assess the research preview's capabilities and limitations.
  2. Follow the official announcement's Genie 4 waitlist link to register interest.
  3. If granted access, describe a world and use the provided navigation and camera controls to explore it.
  4. Steer the scene with short natural-language changes and assess its visual, audio, and spatial consistency.

Genie 4 pricing

Not confirmed

Genie 4-specific pricing has not been announced. Google lists Project Genie within Google AI Ultra benefits, but Genie 4 remains a restricted research preview with broader access announced as coming soon.

Check current pricing ↗

What to know before you use it

  • Access is restricted to trusted testers. Announced Project Genie, Gemini API, and Google AI Studio routes are coming soon; joining the waitlist does not guarantee admission.
  • Continuous interaction is limited to a few minutes. Visual details and object permanence can drift, particularly during longer sessions, repeated complex edits, and navigation through a scene.
  • Complex physics, spatial accuracy, legible text, real-time audio accuracy, and long-term consistency remain research challenges.
  • Complex interactions among users and multiple independent agents remain a research challenge. The model cannot simulate real-world locations with perfect accuracy.

Frequently asked questions

Can I use Genie 4 immediately with Google AI Ultra?

Current Genie 4 access is limited to trusted testers. Ultra access through Project Genie is announced as coming soon, so a subscription alone does not provide Genie 4 access.

Does Genie 4 generate sound?

Yes. It generates native audio on the fly, including ambient sounds, music, and interactive dialogue.

Is it suitable for precise physical simulation?

It remains an exploratory research model. Physics, spatial accuracy, and long-term consistency are active research areas, so its scenes should not be treated as validated physical simulations.