Based on the Genie 3 world model launched by Google DeepMind last year, Google announced an experimental program called "Project Genie" that allows users to experience the power of the Genie 3 world model firsthand and generate their own interactive virtual world.
However, the service currently has a very high barrier to entry, and is only open to US users who have subscribed to the Google AI Ultra plan (which costs as much as $250 per month, or about NT$8000) to apply for a trial.
Powered by Nano Banana Pro, create an interactive world with just one sentence.
At the heart of Project Genie is the Genie 3 world model developed by DeepMind. Simply put, it's not a traditional game engine, but an AI system that can "understand" and "predict" how the visuals will change in response to user actions.
The current demo version offers three modes: "World Sketching," "Explore," and "Remixing." The gameplay is quite interesting:
• AI drawing base:The system will first use the Nano Banana Pro model (Google's advanced image generation model) to draw a "source image" based on the text prompts.
• Setting parameters:You can decide the protagonist's appearance, perspective (first-person, third-person, or isometric), and exploration method.
• Generation and Interaction:Genie 3 will take over this image and transform it into a dynamic, interactive virtual environment.
It's not a game engine, so there are still many limitations.
Although it looks like a game, Google specifically emphasizes that Genie 3 is not a "game engine." This means that it lacks traditional game mechanics (such as scoring, health, or mission objectives), and purely simulates physical interactions and visual feedback.
In addition, the current technology demonstration still has many hardware limitations:
• Duration limit:Each generated interactive experience has a maximum duration of 60 seconds.
• Image quality limitations:The resolution was locked at 720p, and the frame rate was only 24fps.
Showing off the muscles of the "world model"
While the image quality and duration are limited, this is a rare opportunity for Google AI Ultra subscribers to experience DeepMind's cutting-edge technology. Genie 3 demonstrates that AI is no longer just about statically generating images, but can understand physical space and causal relationships (e.g., pressing the right mouse button moves the character to the right, and the background moves accordingly). This will be very useful for training AI agents to understand the real world in the future.
Analysis of viewpoints
"World modeling" is currently one of the most popular "holy grails" in the field of AI, and Project Genie means that developers are one step closer to the future of "AI-generated games in real time." For example, simply inputting "I want to play a parkour game in a cyberpunk city filled with lava" will allow AI to automatically generate and render playable visuals without needing to download tens of gigabytes of game assets beforehand.
Although Project Genie is currently in the expensive technology demonstration phase, the computational logic revolution it represents may completely revolutionize the game development and simulation training industry over the next 10 years.
