Google DeepMind发布世界模型Genie 4：实时生成带原生音频的互动世界，先向受信任测试者开放研究预览
Google旗下研究机构Google DeepMind于10月27日发布新一代世界模型Genie 4，用户设定一个事件后，模型可实时模拟世界如何回应、发出什么声音、人物如何移动、反应和说话；Genie 4已以研究预览形式向受信任测试者开放，后续将面向Google AI Ultra订阅用户（通过Project Genie）以及Gemini API与Google AI Studio的早期合作伙伴推出。
DeepMind称Genie 4在四个方面取得进展。第一，实时塑形：用户可用自然语言在模拟运行中修改天气、移动镜头或引入新的人物与物体，Gemini会把提示适配到场景语境，并在用户两次输入之间持续观察、引导场景连贯推进。第二，物理与人物真实感：模型对空气动力学、碰撞等真实世界物理的理解更强，人物动作更自然，并可被实时互动操控。第三，原生音频：Genie 4是DeepMind首个原生生成音频的实时世界模型，每个世界都带其环境音，从林间风声到海滩浪声，并引入可随输入与事件即时调整的互动对话。第四，复杂事件推理：Gemini 3.8 Flash在循环中担任引导智能体，持续观察场景、把目标拆解为步骤、审查生成的画面以核对每一步是否完成；在与Gemini 4 Argon及Gemini Robotics的实验中，系统把用户指令拆解为步骤，实时演算了Apptronik的Apollo人形机器人下跳棋的过程。
据DeepMind介绍，创意产业、新闻、教育、游戏与工程领域的测试者已在试用：电影人把Genie 4当作预演的「草稿模式」，实时调整镜头、灯光、对话与动作；摄影记者探索把一张二维档案照片在数秒内变成可进入的模拟环境；游戏开发者用自然语言在写引擎代码之前生成可探索的原型；企业伙伴则评估其快速生成场景的能力。DeepMind称，理解与模拟世界是迈向通用超级智能（AGI）的关键一步，这类模拟可为机器人和智能体提供安全的训练环境。
DeepMind同时说明了局限与安全措施：Genie 4仍属早期实验模型，模拟运行越久、经历多次复杂提示编辑或大幅移动视角后，视觉细节与物体恒存性可能偏移；复杂物理互动、实时音频准确性、可读文本渲染与长期一致性仍是活跃研究方向。发布前，DeepMind对模型做了内部严格安全评估，设有安全护栏；所有由Genie 4生成的视频与音频都将带有SynthID数字水印，用于标识生成内容。
Google DeepMind Unveils Genie 4 World Model: Real-Time Interactive Worlds With Native Audio, Rolling Out First as a Research Preview to Trusted Testers
Google DeepMind, Google's research lab, announced its new world model Genie 4 on October 27. A user can set an event in motion, and Genie 4 simulates what happens next: how the world responds, how it sounds, and how the people in it move, react, and speak. Genie 4 has started rolling out as a research preview for trusted testers, and is coming next to Google AI Ultra subscribers in Project Genie and early access partners of the Gemini API and Google AI Studio.
DeepMind said Genie 4 advances four capabilities. First, real-time steering: users can modify a running simulation with natural language — changing the weather, moving the camera, or introducing a new person or object — while Gemini adapts the prompt to the scene and keeps steering the world cohesively between inputs. Second, realism for physics and human action: the model simulates aerodynamics, collisions, and other real-world physics, and its characters move naturally and can be steered interactively in real time. Third, native audio: Genie 4 is DeepMind's first real-time world model to generate audio natively, from ambient sounds such as wind in the trees and waves at the beach to interactive dialogue that adapts on the fly. Fourth, complex event simulation: Gemini 3.8 Flash serves as a guiding agent in the loop, observing the scene, breaking goals into steps, and reviewing frames to verify each step; in experiments bringing together Gemini 4 Argon, Genie 4, and Gemini Robotics, the system broke down a user-instructed task into real-time steps to simulate Apptronik's Apollo humanoid playing a game of checkers.
Trusted testers across creative industries, journalism, education, gaming, and engineering have been trying the model: filmmakers use it as a rapid draft mode for pre-visualization, adjusting camera angles, lighting, dialogue, and action in real time; photojournalists explore turning a single 2D archival photograph into a grounded simulation in seconds; game developers go from an idea to an explorable prototype through natural language before writing engine code; and enterprise partners evaluate rapid scenario generation. DeepMind said understanding and simulating the world is a crucial step toward general super intelligence (AGI), and that such simulations can provide safe training environments for robots and agents.
DeepMind also detailed the model's limits and safeguards: Genie 4 remains an early, experimental model, and visual details and object permanence can shift or drift over time, especially in longer runs, after multiple complex prompt edits, or while navigating a scene; complex physical interactions, real-time audio accuracy, legible text, and long-term consistency are active research areas. Before the preview launch, DeepMind ran rigorous internal safety evaluations and built in safety guardrails, and all video and audio generated by Genie 4 will carry SynthID, its imperceptible digital watermark for identifying generated content.
来源：GSM·美国（原文链接）
图片来源：图：Google公司标志。图片：Wikimedia Commons，公共领域（文字与简单图形标志）。 / Photo: The Google logo. Source: Wikimedia Commons, public domain (text and simple geometric logo).
