Google Launches Genie 4 World Model: Generates 2-Minute Interactive Scenes, Supports Real-Time Natural Language Edits

On October 26, 2026 (local time), Google officially released the fourth-generation Genie world model, Genie 4, with details published on the DeepMind official blog.

Genie falls into the category of AI world models. Unlike one-way playback videos generated by ordinary AI, it produces interactive virtual environments where user actions and input commands trigger corresponding feedback from the environment—this is a popular research and development direction in the current AI industry.

According to official statements, this generation has four key core updates:

1. Higher image fidelity of generated virtual scenes
2. Adjusting current experience content in real time with natural language prompts, no need to regenerate the entire segment
3. Added native audio generation capability, no need for additional audio tools
4. Powered by the Gemini 3.8 Flash model to support the computing demands of the simulation process

In terms of duration, the previous generation Genie 3 supported a maximum of 1 minute of interactive experiences, while Genie 4 has raised the time limit to 2 minutes.

Google currently plans three types of application scenarios:
Film and television teams can use it for pre-visualization, quickly validating shot ideas without shooting physical samples.
Game developers can quickly build gameplay prototypes, avoiding making art assets and debugging physics rules from scratch to verify gameplay feasibility at low cost.
Robot R&D teams can use it to generate diverse, cost-intensive rare training scenarios, reducing investment and risks of real-environment training.

As of press time, the official has not released details such as Genie 4’s open trial time and access thresholds. The source of this information is a report from The Verge.