Google’s Genie 4 turns text prompts into interactive worlds with sound

Google has introduced Genie 4, a new AI world model that generates interactive simulations with visuals, movement, sound, and dialogue in real time. The company is initially making the system available as a research preview for selected trusted testers.

Genie 4 builds on the earlier Genie 3 model and extends the length of generated experiences to roughly two minutes, Jay Peters reports for The Verge. Google says the model can create photorealistic scenes that users navigate with keyboard controls while changing elements through text prompts.

A user could, for example, alter the weather, add an object, change a camera position, or introduce a new character. According to Google, the system adapts those changes to the existing scene rather than generating a separate clip.

Audio, dialogue, and planning

One of Genie 4’s major additions is native audio generation. Google says the model produces ambient sounds such as wind, waves, and environmental noise as part of the simulation. Characters can also respond through interactive dialogue that changes as users act or issue new prompts.

The model combines Genie’s visual simulation with Gemini models. Gemini 3.8 Flash acts as a guiding system that monitors the generated scene, breaks a goal into steps, and checks whether each step has been completed. Google says this could help create more coherent multi step scenarios for robotics training, game prototypes, filmmaking previsualization, and other creative work.

The company also describes tests involving Gemini 4 Argon, which can instruct Genie 4 to simulate tasks in sequence, including a humanoid robot playing checkers.

Google acknowledges significant limits. Objects and visual details can drift during longer sessions or after several complex edits. The company also lists physical interactions, readable text, spatial precision, audio accuracy, and long term consistency as active research areas. Generated video and audio will include Google’s SynthID watermark, the company says.

Google plans to expand access to AI Ultra subscribers through Project Genie, as well as selected Gemini API and Google AI Studio partners.

Sources

• Genie 4 – Google DeepMind
• Google launches Genie 4, a world model powered by Gemini 3.8 Flash designed to generate interactive AI simulations, initially as a preview to trusted testers – The Verge