Google DeepMind 发布 Genie 4 实时世界模型，由 Gemini 负责规划
Google Deepmind's Genie 4 puts a modern spin on agent designs that predate LLMs
AI 导读
Google DeepMind 发布 Genie 4，这是 Genie 3 的后继版本，能逐帧计算场景并实时响应用户输入，用户可用键盘移动、用文本提示改变事件走向。
摘要与补充说明
DeepMind 称它是公司首个直接生成声音的实时世界模型，每个世界带有自己的背景音，角色之间可以对话并实时反应。Gemini 模型在后台负责规划，例如 Gemini 3.8 Flash 持续观察画面、把目标拆成步骤并检查每步是否完成；DeepMind 还在测试 Gemini 4 Argon，把 Genie 与 Gemini Robotics 的机器人研究连接起来，演示中模拟了 Apptronik 人形机器人 Apollo 下棋。
正文
Genie 4 is supposed to simulate interactive worlds in real time, including sound and talking characters. Gemini models plan in the background what happens next in the simulation. Only select testers can try the world model for now.
Google Deepmind has introduced Genie 4, the successor to Genie 3, which the company showed in August 2025. The world model doesn't produce finished videos but calculates a scene frame by frame, reacting to input as it goes. Users move through the scene with the keyboard and steer events with text prompts.
According to Deepmind, Genie 4 is the company's first real-time world model that generates sound directly. The company says each world has its own background audio, such as wind in the trees. Genie 4 also handles dialogue, with characters that are supposed to talk to each other and react in real time to user input or events.
Deepmind sees the biggest gain in realism in how the model renders people. Characters are supposed to move naturally and react to their surroundings, and users should be able to control them directly.
Gemini models plan ahead to keep the world coherent
According to Deepmind, Genie 4 builds on capabilities of the video model Gemini Omni so the world looks and behaves realistically. Other Gemini models help plan the simulation to keep it consistent.
If a user changes the weather with a text prompt, Gemini is supposed to adapt the instruction to the running scene. Deepmind says Gemini also steers events between inputs so the world keeps developing in a way that makes sense.
Deepmind explains the process with multistep tasks. Gemini 3.8 Flash watches the scene continuously, breaks a goal into steps, and uses new frames to check whether each step worked. If the world changes or the user steps in, the model adjusts its plan.
Deepmind is also testing Gemini 4 Argon to tie its Genie work to its robotics research, Gemini Robotics. In one demo, Argon breaks the task of playing checkers into steps. Genie 4 gets these as text instructions and uses them to simulate Apptronik's humanoid robot Apollo playing the game, with no real robot involved.
Longer sessions still trip up the model
Deepmind calls Genie 4 an early, experimental model. According to the company, the world stays consistent over short sessions, but visual details can shift and objects don't always persist in longer simulations or after several complex prompt changes.
Deepmind says it's still researching complex physical interactions, accurate real-time sound, and readable text.
For now, Genie 4 is only available to select testers as a research preview. It's set to reach Google AI Ultra subscribers in the Project Genie web app soon, along with partners who have early access to the Gemini API and Google AI Studio.
Google says all videos and audio tracks will carry its SynthID watermark, which invisibly marks AI-generated content.
Once they mature, world models could serve as training environments where robots and AI agents learn from simulated experience rather than just from data. Deepmind sees pairing Gemini with Genie as an early step toward hybrid world models that combine the reasoning of language and code models with pixel-level simulation.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.