Google DeepMind unveiled Genie 4 on 27 October, a model that generates interactive, photorealistic worlds in real time. Alongside the visuals it produces audio, including ambient sound, music and character dialogue. Users can change the weather, move the camera or add people and objects with plain text instructions while the simulation runs. The company says a scene stays coherent for up to about two minutes, twice as long as Genie 3. Access starts with selected testers and will widen to Google AI Ultra subscribers and Gemini API partners.

What it means: World models, systems that simulate an environment and respond to actions, are becoming a core tool for training robots and AI agents. Instead of testing a system only in the real world, which is slow and expensive, it can be trained on thousands of artificial scenarios. Adding sound and the ability to make changes through plain text brings Genie 4 closer to a content production tool, for example for game prototypes, training material or virtual tours. Some caution is needed, though: this is a research preview, DeepMind itself admits that objects, audio and text can drift over longer sessions, and there is no business pricing or availability yet. For Greek game studios, education technology firms or tourism marketers, the practical step today is to watch the API and experiment once it opens, without building production work on it. It is also worth remembering that the output carries a SynthID watermark and that the EU AI Act requires clear labelling of synthetic content when it is published, so anyone using it in advertising or customer material will need to disclose it.

Read the original on Google DeepMind

Related service: Custom software

Related stories