Google introduces Genie 4: interactive AI worlds with sound and real-time control

Image: official Google DeepMind demonstration thumbnail, showing a model-generated environment.
Прочетете статията на български
Google DeepMind has introduced Genie 4, a model that generates interactive environments and their sound together. Users can steer a scene with text as it unfolds. The official demonstration is embedded below.
Google announced the model on 27 October 2026. This article was published on 28 October.
What the new model demonstrates
According to DeepMind, Genie 4 lets users change weather, camera movement, characters and events through natural language. Generated audio includes ambient sound, music and interactive dialogue. The model draws on the Gemini family to support environmental behaviour and continuity.
Watch the official demonstration
Official Genie 4 demonstration by Google DeepMind, showing model-generated environments. Watch on YouTube.
Who can use it
As of 28 October, access is limited to trusted testers through a research preview. Google AI Ultra subscribers using Project Genie and selected Gemini API and AI Studio partners are the next announced groups, with no specific start date.
The product page describes interactions lasting a few minutes and remaining limitations in object persistence, spatial detail and audio.
The TechWork.bg perspective
For software engineers, the interesting question is how such an environment can be controlled, observed and evaluated. A prototype may look convincing, while a product still needs repeatable scenarios, measurable errors and clear behavioural boundaries.
We see scope for experiments with interactive interfaces and agent training. Engineering teams should first establish whether the simulation preserves the relationships their application depends on before using its outputs for decisions with real-world consequences.
Sources
Google DeepMind · Google · Genie — capabilities and limitations