ReadingResearchRadarInvestment frameworkAI Company Primary Sources
Sign in / Sign up
← Back to reading
TopicDisconfirmAI EditorialUpdated 2026-10-28 13:10

谷歌Genie 4模型

3 steps3 pieces3 Sources

美国谷歌公司旗下DeepMind部门于2026年10月27日正式发布新一代世界模型Genie 4。该模型实现了原生多模态生成,支持在实时交互模拟中同步生成音频(包括环境音、音乐和对话),并将单次模拟时长从上一代Genie 3的一分钟延长至两分钟。Genie 4深度整合了Gemini系列模型(如Gemini Omni和Argon),允许用户通过自然语言指令实时修改场景元素(如天气、摄像机角度或新增角色)。目前,该模型仅向受信任测试者开放研究预览,计划随后向Google AI Ultra订阅用户及API早期合作伙伴推出。尽管谷歌强调其在机器人训练环境和影视预可视化方面的潜力,但团队承认随着模拟时间延长,物体持久性和细节仍会出现漂移,且音频准确性和画面文字可读性尚待优化。

How it got here

  1. 2025-08-01

    谷歌发布前代模型Genie 3的研究预览版,当时模拟时长限制为60秒。

    techcrunch.com ↗
  2. 2026-08-01

    AI视频初创公司Runway发布其第二代世界模型GWM Worlds 2,具备动态响应和音频生成功能,加剧了世界模型领域的竞争。

    techcrunch.com ↗
  3. 2026-10-27

    谷歌DeepMind正式宣布推出Genie 4,支持原生音频生成、实时自然语言交互及两分钟模拟时长,面向受信任测试者开放。

    deepmind.google ↗

What is still unsettled

  • Genie 4何时正式向公众全面开放,而非仅限于AI Ultra订阅者和API合作伙伴?
  • 谷歌如何解决长时间模拟中的物体持久性漂移和音频准确性问题?
  • 在世界模型领域并购潮(如AMD收购World Labs)背景下,Genie 4的商业化路径将如何调整?

Sources

Sign in / Sign up
ReadingResearchRadarInvestment frameworkAI Company Primary Sources
deepmind.google
Google DeepMind官方博客
↗
  • techcrunch.comTechCrunch报道↗
  • theverge.comThe Verge报道↗
  • Coverage in this topic

    • Google DeepMind Launches Genie 4 World Model with Native Audio and Real-Time Steering2026-10-28News

      Google DeepMind introduces Genie 4, featuring synchronized audio generation, real-time scene control via natural language, and extended simulation durations.

    • Google Announces Genie 4 World Model: Adds Audio, Real-Time Prompt Updates, Doubles Simulation Length to Two Minutes2026-10-28News

      Google launches Genie 4, adding audio generation and allowing real-time experience updates via natural language prompts, while extending simulation duration from one minute to two.

    • Google DeepMind Announces Genie 4: A New Generation of Natively Multimodal World Simulation2026-10-28News

      Google DeepMind has unveiled Genie 4, a new world model enabling real-time interactive simulation with native audio and dialogue generation, currently in research preview.