Google DeepMind's Genie 4 adds native audio to world models

Google DeepMind introduced Genie 4 on 27 October. The Genie series turns a prompt or image into explorable 3D worlds that react in real time. The headline feature is native audio: ambient sound, music and interactive dialogue are generated on the fly. It is a research preview for Google AI Ultra subscribers in Project Genie and early-access Gemini API and AI Studio partners.

Users can change the weather, move the camera or add people and objects in plain language. TechCrunch says scenes stay consistent for up to two minutes, double the previous limit. Secondary reports say Gemini 3.8 Flash acts as an in-loop agent that plans steps and checks frames; DeepMind's own text says Gemini models provide reasoning and Gemini Omni anchors appearance and behaviour. Output is reportedly tagged with SynthID.

Earlier Genie versions were silent. Generating sound with the image means footsteps match steps and a door creaks as it opens, which makes the experience feel far more real.

Details drift over longer runs, and accurate audio and legible in-world text remain unsolved, per TechCrunch.

It is not ready to replace a game engine, but it makes an early concept explorable in minutes.

A world model predicts how an environment looks and changes after an action, reacting to user input in real time. Such models also train robots in simulation. A research preview is a limited trial before stability and pricing are set.

The same week Higgsfield launched Playable, which turns a prompt into a playable mobile game or ad, and xAI released the first seven minutes of a Grok-made film. Playable formats may start to rival video. Wider access and API pricing for Genie 4 are not yet announced.

• Prototype faster: watch these tools for explorable store, product or training drafts.
• Consider playable ads where attention is short.
• Check watermarking and commercial terms before publishing.