Google opens Genie 4 research preview for trusted testers

Top News · 809 HN points

Genie 4 is Google’s real-time, interactive world model, now in a research preview for trusted testers. You can steer a running simulation with natural language, with Gemini adapting your prompts to the scene and monitoring what happens between inputs.

Here’s what you can do with it:

• Change a world while it runs, with enhanced photorealism and fluid movement.
• Hear native ambient audio and interactive dialogue that responds to your input or events in the world.
• Gemini 3.8 Flash guides complex multi-step scenarios by observing frames, planning steps, checking progress, and adapting.
• Google plans access for Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio.

One catch: Environments can drift during longer sessions, complex edits, or navigation. Spatial accuracy, legible text, audio accuracy, and long-term consistency remain research problems.

Sources: deepmind.google