Back to AI game news
AI game news

Google opens Genie 4 to developers through the Gemini API for interactive world prototypes

By GameSlash

The November 19 announcement brings Genie 4's 720p, 24 fps promptable worlds and native audio to Gemini API and AI Studio developers.

On November 19, 2026, Google announced that Genie 4 is accessible to developers through the Gemini API (Interactions API) and Google AI Studio. This is a separate access expansion following the November 18 launch of Genie 4 in Project Genie for eligible Ultra subscribers. Developers can now build experiments around interactive AI-generated environments, instead of only trying them in the consumer-facing Project Genie interface.

Official still from Google's Genie 4 interactive world-generation demonstration
Image: Google DeepMind / official Genie 4 demonstration — visual reference from Google's November 18 launch, illustrating generated interactive worlds.

What developers can actually control

According to Google, a world can start from a written prompt, a reference image, or both. Genie 4 produces continuous 720p visuals at 24 fps with native synchronized audio, including environmental sounds, effects and character dialogue. Keyboard controls let users move through the environment, while prompts can change weather, lighting, objects and events during a session.

This matters for game prototyping: creators could try the mood of a scene, a change in weather or an interactive spatial idea before committing to assets and level implementation. It does not mean Genie 4 outputs a ready-to-ship Unity or Godot project, complete game logic, or editable meshes; these are world simulations, not conventional engine builds.

Different frame from an official Google Genie 4 demonstration showing generated world visuals
Image: Google DeepMind / Genie 4 launch — a second official demo still; not a GameSlash-generated gameplay screenshot.

How to explore the developer release

Google points developers to AI Studio and its Genie documentation to inspect starter apps and access details. Start with a reference scene or text prompt, test interactive movement, change one feature with a live prompt, and examine visual continuity and audio behavior. Test restrictions, pricing and API availability against the live documentation before planning production use.

Google says every generated output receives an imperceptible SynthID watermark. Its announcement makes no claim that the output can be exported as traditional game assets; treat this as an interactive previs and experimental environment-generation capability. Original announcement.

View source