Google DeepMind introduced Genie 4 on Tuesday as an early research version for trusted testers. The AI model creates interactive virtual worlds as users explore them, rather than producing a finished video. Google says it offers more realistic people and movement, together with music, background noise and dialogue.
Users can type instructions during a simulation to change the weather, adjust the camera or add people and objects. Google’s Gemini 3.8 Flash helps guide the experience by planning steps and checking the images it generates. Scenes can remain consistent for about two minutes, twice the limit of the previous model, Genie 3.
Potential uses include planning films, trying out game ideas and creating training situations for robots. However, the simulations still have problems: objects can change position unexpectedly, and their movement does not always follow real-world rules. Google also acknowledges difficulties with accurate sound and readable text inside the generated worlds.
For now, access is restricted to trusted testers. Google plans to bring Genie 4 to Google AI Ultra subscribers through Project Genie in the coming weeks. Selected early partners will also receive access through the Gemini API and Google AI Studio, tools through which developers can use Google’s AI models.
