• Source:JND
Add as a preferred

Google DeepMind has introduced Genie 3, the latest version of its AI “world” model capable of building immersive 3D environments where both humans and AI agents can interact in real time. Unlike traditional game engines that rely on manually created assets, Genie 3 generates entire virtual spaces from prompts—automatically and dynamically.

The idea behind world models like Genie is to enable AI systems to understand and simulate environments—useful not just for gaming or entertainment, but also for training other AI systems or robots in realistic scenarios.

ALSO READ: Microsoft’s 2025 Layoffs Cross 15,000, Second Largest Since 2014

Genie 3 brings a big leap forward in realism and memory. According to DeepMind, users can now stay in these AI-generated worlds for a few minutes, a significant upgrade from the 10 to 20 seconds possible in Genie 2. More importantly, the system now remembers what it generates—visual details like writing on a chalkboard or colours on a wall will remain intact even if you look away and return later. The model runs these scenes at 720p resolution and a smooth 24 frames per second.

Another key improvement is the introduction of “promptable world events”. This feature lets users change things on the fly—like making it rain, summoning characters, or adjusting the environment—simply by typing a command.

When Genie 2 launched in December, it was notable for being able to create interactive spaces from a single image. But its limitations were clear: short-lived scenes and visuals that would sometimes warp or shift unpredictably. Those issues are addressed in Genie 3, which now keeps environments more stable and persistent.

However, don’t expect to try Genie 3 just yet. Google DeepMind is limiting access to a small group of researchers and creators for now. The rollout is part of a cautious research preview, aimed at identifying safety concerns and refining controls. Most interaction features will be restricted initially, and any readable text in the world will only appear if it's part of the original prompt.

ALSO READ: Group Added By Unknown Contact? WhatsApp Now Gives You A Safety Check

The team leading the project includes a former co-lead of OpenAI’s Sora video generation initiative, adding even more AI firepower to DeepMind’s virtual world ambitions.