Genie 3 is the newest iteration of Google DeepMind’s world model AI, a technology that can generate interactive 3D environments on-the-fly from simple text prompts. It allows users to navigate and interact with AI-generated worlds in real time at 24 frames per second, with 720p resolution, maintaining coherence for several minutes.

Previously, environments lasted only 10 to 20 seconds. Genie 3 now sustains interactivity for multiple minutes. It remembers object placements and environmental changes, even if you look away and return later. Users can modify scenes in real time, adjusting weather, adding characters, or altering settings via text.

Read more: Best 10 Smartphones Releasing in August 2025

Genie 3 provides richly simulated environments ideal for embodied agent research, teaching AI systems how actions translate into outcomes in virtual settings. It also offers creative potential for game designers, educators, and content creators by enabling dynamic, evolving world-building. DeepMind sees Genie 3 as a foundational step toward artificial general intelligence (AGI), offering agents a limitless training curriculum in simulated worlds.

While impressive, Genie 3 still has constraints. The range of agent actions (“action space”) is somewhat limited. Difficulty simulating complex multi-agent interactions and real-world geographic accuracy. Text rendering can be poor unless explicitly included in the prompt. Sessions are still only a few minutes, not yet suitable for hours-long experiences.

Read more: Speaking with AI: Best Free Apps for English Speaking Practice

Genie 3 is currently available as a limited research preview, accessible only to select academics and creators. DeepMind emphasises careful, responsible rollout to monitor risks and gather feedback.