Genie 4 - Gemini API - interactive worlds - AI development
3 min lectura

Genie 4 API: Price and Time Limit for an Interactive World

Google opened Genie 4 in the Gemini API. Understand per-second cost, the 120-second cap, session cleanup, and a bounded developer experiment.

Genie 4 API: Price and Time Limit for an Interactive World

Suppose you want a visitor to explore a spatial concept during a product demo rather than watch a fixed video. On November 19, 2026, Google announced Genie 4 access through the Gemini Interactions API and Google AI Studio. This gives developers a different path from using the Project Genie app: create a live session, stream an interactive world, and send movement or scene changes from your product. The access and cost shape the experiment. Google's developer documentation requires a paid-tier Gemini API key and describes a 120-second hard cap.

What you can build with the API

Genie 4 generates a continuous 1280 × 720 world at 24 frames per second with stereo audio. You can seed it with text, a 16:9 image, or both. During the session, keyboard input such as WASD or arrow keys steers movement, while text events can change conditions in the world. Google recommends WebRTC for a browser client and also documents WebSocket connections. The launch says outputs carry an imperceptible SynthID watermark.

That is useful for testing the feel of an interactive art direction, showing a walkable concept to a client, or trying a scene before building a conventional game. It is not a persistent 3D engine. Google documents object drift when a user looks away and visual degradation in longer sessions. The API does not support live microphone or webcam input, function calling, portrait 9:16 output, or resuming a completed world session.

Price is part of the product decision

Google bills 240 video output tokens per frame, or 5,760 per second at 24 fps. At its published Standard rate, that is approximately $0.10 per second, or about $6.05 per minute, with stereo audio included. A session that reaches the two-minute cap would be about $12.10 in video output. This is arithmetic from the current price page, not an invoice we measured.

One operational detail could matter more than a clever prompt: closing a browser tab or leaving a local code block may detach the connection without immediately stopping generation on Google's server. The documentation says to call complete() to end the session and stop billing. Add a timer, a per-session budget, and a cleanup path when a connection fails before offering a public demo.

Run one bounded experiment

  1. Write one measurable scene brief, such as “Let a visitor walk through an abstract room and change the lighting from day to night.” Avoid recognizable people or copyrighted brand worlds that may trigger safety filters.
  2. Budget the run first. Twenty seconds are about $2.02 in output at the published rate. Decide how many attempts you can afford and set a hard limit for each session.
  3. With a paid API key, follow the documented lifecycle: create an interaction using genie-4, connect through WebRTC or WebSocket, send world_setup, try one movement and one text change, then call complete().
  4. Record time to first frame, response to keyboard movement, and whether objects remain stable after a turn. Reuse the same brief to compare attempts.

Google says serving regions in the United States, Western Europe, and Japan offer the best responsiveness; users farther away may see slower steering. If your audience is in Latin America, measure that network delay before you plan a reflex-heavy experience. A two-minute prototype also tells you little about long-term retention or the economics of a finished game.

Primary sources: Google's launch post (November 19, 2026), the Genie 4 API guide, and official pricing. SoyJB did not run a billable session; the proposed test requires a paid-tier key.

Comentarios

Aún no hay comentarios. Sé el primero.

Deja un comentario