Google Releases Next-Generation World Model Genie 4

en.Wedoany.com Reported - On October 27, Google announced the launch of its next-generation AI world model Genie 4, opening a research preview to select trusted testers. The model supports highly realistic real-time interactive simulation, synchronously generating ambient sound effects, music, and character dialogue, and allows users to adjust scenes, weather, camera angles, and character actions in real time through natural language. Google plans to subsequently open it to the Google AI Ultra subscription service and developer platforms.

Genie 4 integrates the multimodal processing and reasoning capabilities of the Gemini series models, with Gemini Omni responsible for supporting the realistic appearance and behavioral performance of simulated scenes, while Gemini 3.8 Flash coordinates real-time simulation of multi-step events by continuously observing frames, decomposing tasks, and checking execution results. The model can generate dynamic scenes involving aerodynamics, object collisions, and character interactions, and continuously changes environmental states according to user instructions.

Google also showcased joint research results combining Genie 4 with Gemini Robotics. In one experiment, Gemini 4 Argon decomposed the actions required for a robot to perform a task into multiple steps, and Genie 4 then generated corresponding simulated frames in real time. The demonstration included the Apptronik Apollo humanoid robot performing checkers-related operations.

Currently, Google has organized testers in the film and television, news, gaming, education, and engineering fields to conduct application validation, covering film and television scene previsualization, interactive game prototypes, and robot training environment generation. Subsequent access will cover Google AI Ultra subscribers in Project Genie, as well as early partners of the Gemini API and Google AI Studio.