Genie 4: Google’s AI Now Builds Hyper-Realistic Worlds in Real Time

Google has taken a significant new step toward creating hyper-realistic interactive worlds with Genie 4. The latest update to the artificial intelligence model arrives with substantial improvements, marking a noticeable leap in quality compared to the previous version introduced just over a year ago as part of the company’s broader push into generative world simulation.

The first major optimization Genie 4 brings over its predecessor lies in human simulation. It is now possible to generate simulations featuring people who look significantly more realistic than before. Furthermore, the AI demonstrates a better physical understanding of the world, allowing it to create scenarios where outcomes closely mirror real life when dealing with collisions, aerodynamics, and other complex environmental dynamics.

Another key addition in Genie 4 is native audio generation. This means the artificial intelligence supports the creation of dialogue, ambient sound, effects, and music in real time. What Google has developed is highly promising, building upon the foundation of other models already available within its ecosystem, such as Gemini 3.8 Flash and Gemini Omni.

While previous generations of the model appeared geared toward generating interactive environments similar to video games, Genie 4 elevates the concept to another level. The new AI offers the ability to create situations that naturally trigger subsequent actions, and everything a user imagines can be added or modified on the fly through a text prompt.

Google maintains that Genie 4 can serve practical professional purposes, such as allowing a filmmaker to generate a rough draft of a scene being prepared for a series or movie. For instance, users can adjust lighting, camera angles, or action in real time. The company also notes that video game studios could use the tool to prototype design concepts before committing to full development, and even mentions the option of creating interactive environments from a single existing static image.

Naturally, Genie 4 still has limitations. Google warns that the model cannot yet maintain consistency across certain elements when a simulation becomes too large or runs for an extended period. The same issue arises when multiple modifications are made simultaneously, particularly if they involve a high level of complexity.

“Complex physical interactions, real-time audio accuracy, legible text rendering, spatial precision, and long-term coherence are areas of active research,” the California-based company cautioned in this regard. Consequently, users should not expect anyone to attempt to build a complete film or fully realized video game solely using Genie 4 at this stage.

For now, Genie 4 is limited to a trusted group of testers, but Google intends to open access to the general public soon. However, users should prepare to pay for the privilege. Initially, it will only be offered to subscribers of Google AI Ultra, the premium plan priced at up to 219.99 euros per month.