Google 向 Gemini API 开发者开放 Genie 4 世界生成能力
Google Opens Genie 4 World Generation to Gemini API Developers
Google 于 11 月 19 日通过 Gemini API 和 Google AI Studio 开放 Genie 4,可生成支持键盘控制与实时文本提示的连续 720p、24 fps 交互视频及原生同步音频。该模型仅面向付费层,视频输出约为 $0.10 per second,单次会话最长 120 秒,并可通过 WebRTC 或 WebSocket 接入。其场景稳定性可能在 60 至 90 秒后下降,已结束的交互无法恢复。
推荐理由
精选AI 评分 83
文章汇总了接入方式、计费、延迟与连续性限制,可帮助开发者评估短时交互世界是否适合现有应用。
主题
正文
Google made Genie 4 available through the Gemini API and Google AI Studio on November 19, 2026, giving developers access to interactive worlds that respond to keyboard controls and live text prompts. The model generates continuous 720p video at 24 frames per second with synchronized native audio.
The developer release uses models/genie-4 on the v1beta Interactions API. A world can be seeded with text, a single reference image, or both; WASD and arrow keys control movement and camera orientation, while text events modify the scene or supply character dialogue. The API documentation distinguishes this responsive frame rendering from offline generation of a fixed video clip.
Google lists prototyping interactive experiences, visualizing spatial concepts, and training robotics and AI agents as possible uses, and says every output carries imperceptible SynthID watermarking.
Paid Sessions and Billing
Genie 4 requires a paid-tier Gemini API key and is not offered on the free tier. Its standard pricing is $1.50 per million tokens for text or reference-image input, $9 per million for text output, and $17.50 per million for video output with stereo audio included. At 240 video output tokens per frame, or 5,760 tokens per second, the listed effective rates are approximately $0.10 per second and $6.05 per minute. Batch API, Flex inference and Priority inference are unsupported.
To build a session, developers create an interaction, connect over WebRTC or WebSocket, and send one world_setup event before steering with user_action or text input events. Browser applications can use an ephemeral token scoped to one interaction, keeping the long-lived API key on the backend. WebRTC is recommended for interactive browser apps, while WebSocket supports Python and Node.js workflows, frame capture and custom pipelines.
Closing a browser tab or detaching the local transport does not immediately stop server generation. Calling complete() ends generation and billing, while cancel() aborts server-side generation, making explicit shutdown part of cost control.
Session Limits and Latency
Across its model documentation and implementation guide, Google caps sessions at 120 seconds, identifies the first 45 to 90 seconds as the strongest period for visual coherence and object permanence, and warns that visual stability may degrade after 60 to 90 seconds.
The guide describes generation over a rolling temporal window without a permanent 3D scene. Looking away from an object for eight seconds or more, or turning 360 degrees, can make landmarks, doors and buttons move or reset. For robotics or agent experiments, the generated environment therefore cannot be assumed to provide a stable long-term map.
Output is fixed at a 16:9 aspect ratio, with no portrait mode or configurable frame rate, and sessions are served only from US data centers. Google’s implementation guide reports a visual response of about 0.2 to 0.5 seconds for native keyboard actions at low network latency, compared with roughly one to two seconds for mid-session text changes. Round-trip network latency above 150 milliseconds noticeably delays keyboard steering.
Genie 4 does not accept live microphone or webcam streams. It also lacks function calling, Google Search grounding, structured outputs and the ability to resume a completed interaction.
Usage Quotas
Genie 4’s rate limits vary by paid project usage tier:
Usage tier Concurrent active sessions Requests per day
Real-time control messages within a session do not consume separate per-minute or per-day request quotas.
The API makes Genie 4 a programmable environment for brief, controlled simulations rather than a hosted persistent world. Its central constraint is continuity: applications can generate responsive scenes on demand, but the current design does not provide a world they can retain and revisit across completed sessions.
AI 热点