Google DeepMind Unveils Genie 4 AI for Real-Time 3D World Generation
Translated & summarized from Vesty by baba
Google DeepMind has unveiled Genie 4, an AI system that generates interactive, photorealistic 3D environments in real-time with simulated physics and voice dialogue. The technology allows users to modify these spaces via text commands and is being developed amidst fierce competition, including from Israeli startup Decart, which is reportedly in acquisition talks with Amazon for $7 billion. Genie 4 aims to advance AI's understanding of the physical world and is seen as a step towards AGI, though it still faces some limitations.
The story in 6 lines · by baba
- Google DeepMind's Genie 4 AI creates interactive, photorealistic 3D worlds in real-time with simulated physics and dialogue.
- Users can modify the generated environments using text commands and keyboard input.
- The system can simulate physical laws for up to two minutes, useful for training robots and specific scenarios.
- Israeli startup Decart, a competitor, is in talks with Amazon for a potential $7 billion acquisition.
- Genie 4 is considered a step towards Artificial General Intelligence (AGI) amid global AI development.
- Access will be provided to Google AI Ultra subscribers and select developers in the coming weeks.
Google DeepMind has introduced Genie 4, a new artificial intelligence system capable of generating photorealistic 3D environments in real-time, complete with voice dialogue and accurate physics simulation. This represents a significant advancement in AI's ability to understand and model the physical world, moving beyond text analysis.
Genie 4, the fourth generation of Google's "World Model," allows users to interact with and modify these virtual spaces using text commands and keyboard input. Unlike traditional video generators, Genie 4 creates dynamic, interactive environments where AI characters respond naturally to changes. The system's capabilities are dependent on available computing resources, but it can simulate physical laws, including aerodynamics and object collisions, for up to two minutes, a doubling of the previous generation's capacity. This duration is sufficient for training autonomous robots or simulating specific scenarios for technical training.
The system employs a hybrid architecture combining Google's language models, such as Gemini Omni, Gemini 3.8 Flash, and Gemini 4 Argon. The language model acts as a "director," breaking down complex tasks, while the world model translates instructions into visuals. An experiment demonstrated its use in training a humanoid robot to play checkers in a virtual setting, testing physical scenarios before real-world implementation.
This development occurs amidst intense global competition in AI. The Israeli startup Decart, founded by Dean Leitersdorf, Orian Leitersdorf, and Moshe Shalev, is a notable competitor. Decart's products, Oasis and Lucy, focus on computational efficiency and real-time editing with low latency, contrasting with Google's emphasis on high-fidelity graphics. Amazon is reportedly in advanced talks to acquire Decart for approximately $7 billion, a move seen as a strategic response to Google and Nvidia's dominance.
Other players in the market include Runway with its GWM Worlds 2 model, AMD's acquisition of World Labs for $8.2 billion, and Rivian's purchase of Third Dimension. The advancement of world models is considered crucial for the development of Artificial General Intelligence (AGI). Despite its progress, DeepMind acknowledges that Genie 4 still faces challenges with visual distortions, text rendering, and object constancy in prolonged simulations. All generated content will be watermarked with SynthID. Access to Genie 4 will be granted to Google AI Ultra subscribers and select developers soon.
Mentioned
The same event, reported separately by each outlet. Open a few to compare what different newsrooms emphasize — and what they leave out.
Centre 1Other 1
