Google DeepMind 发布实时世界模型 Genie 4，支持原生音频
DeepMind 的 Genie 4 能边生成画面边出声音和对话，还能用打字实时改场景，连人形机器人下跳棋都能模拟出来，挺值得玩味。
Google DeepMind 发布 Genie 4，是其首个支持原生音频的实时世界模型，可生成环境音、音乐以及响应用户和场景事件的角色实时对话。用户可以在运行中的模拟里通过输入文字改写场景，比如改变天气、移动镜头或添加人物，由 Gemini 将每次编辑与场景融合并保持连贯。Gemini 3.8 Flash 在循环中充当导演角色，把目标拆解为步骤并校验生成的帧。Google AI Ultra 订阅者和早期 Gemini API 合作伙伴可通过受信任测试者渠道获得访问权。
> Genie 4 is Google DeepMind’s first real-time world model with native audio, generating ambient sound, music and live dialogue from characters who react to users and in-world events.
> Users can rewrite a running simulation by typing, changing weather, moving the camera or adding people, while Gemini fits each edit to the scene and keeps it coherent.
> Gemini 3.8 Flash acts as an in-loop director, splitting goals into steps and checking incoming frames, and Gemini 4 Argon used that loop to simulate Apptronik’s Apollo humanoid playing checkers.
> access starts with trusted testers before Google AI Ultra subscribers and early Gemini API partners. https://t.co/DYN0zqMU97
- theverge16:00原文
