据谷歌官方博客及 The Defiant、Unite.AI 报道,当地时间 11 月 19 日,谷歌正式向开发者开放世界模型 Genie 4(模型标识 models/genie-4),可通过 Gemini API 的 Interactions 多轮交互接口及 Google AI Studio 接入。此前一日已在消费端 Project Genie 向 AI Ultra 订阅用户开放。Per Google's official blog and The Defiant/Unite.AI, on Nov 19 Google opened its Genie 4 world model (models/genie-4) to developers via the Gemini API's Interactions interface and Google AI Studio, a day after granting AI Ultra subscribers access through Project Genie.
Genie 4 支持从文本提示、参考图像或图文结合出发,生成 720p、24fps 的连续可交互世界,并同步生成环境音、音效及角色对白等原生立体声音频,所有输出嵌入 SynthID 隐形水印。开发者可用 WASD 或方向键实时移动,并通过文字指令动态修改天气、光照与场景事件,世界状态由 Gemini 推理保持连贯。生成视频成本约 6.05 美元/分钟。Genie 4 generates continuous 720p/24fps interactive worlds from text or image prompts, with native stereo audio for ambience, sound effects and character dialogue, all watermarked with SynthID. Developers move with WASD/arrow keys and reshape weather, lighting and events via text prompts, with world coherence maintained by Gemini reasoning. Generated video costs about $6.05 per minute.