What is the story
Google's Deepmind team has announced the Genie 3, a cutting-edge artificial intelligence (AI) model that can generate interactive 3D environments from text prompts. The new model allows users and AI agents to interact with these virtual worlds in real time. This is a greater improvement over its predecessor, Genie 2, allowing only short bursts of interaction.
Improved interaction time and visual memory
Genie 3 offers much longer interaction times that last for a few minutes, as opposed to the only seconds that Genie 2 offers. It also boasts visual memory that can hold details for about a minute, ensuring continuity of virtual space. The model generates the world at 720p resolution and 24 frames per frame rate (FPS), providing an immersive experience for the user.
Rapid World Events and Current Limitations
Genie 3 introduces “quickly world events” where users can manipulate elements such as the weather and add characters with simple prompts. However, this model is still a research preview and can only be used to select scholars and creators. Depending on the description of the input world, there are often limitations such as how users interact and easy-to-read text generation.
Potential applications and future access plans
Genie 3's realistic natural scenes, fictional worlds, and ability to create historical simulations will become a valuable tool for researchers to train AI agents. DeepMind has already tested Genie 3 with SIMA generalist agents in various worlds. Despite current restrictions, DeepMind is confirming that ethical protection measures are in place, as they consider ways to expand access beyond the current research preview.
