Vibeleaderboard
← All Intel
Intel / blog

Genie 4 is now available

Source
blog.google
Date

Context

Genie 4, the world model Google DeepMind announced in October with a waitlist, is now usable by eligible Google AI Ultra subscribers through Project Genie, Google's browser app for building and exploring generated worlds. A world model generates an environment you can move through in real time: you steer with the keyboard, and the model produces what you see next in response, so the scene is simulated as you go rather than played back as a finished video. Google AI Pro subscribers can join a waitlist, according to Google's announcement.

The main change for users is that a world can now be edited while you are inside it. With Genie 4 you type a new prompt mid-session, for example to change the weather or add a character, and the world changes without starting over. Genie 4 also generates sound alongside the picture, from ambient noise to character dialogue, which Google says adapts as the scene unfolds. Google's October post explained that a Gemini model acts as the planner behind this, turning a request into actions and tracking what is happening in the scene.

The release also adds a Project Genie , which builds a world's opening look and background music from a conversational description by calling several Google models, and a community gallery where people can publish and remix worlds. The older Genie 3 stays available for longer runs: each generation can now last up to three minutes, up from 60 seconds, and worlds can be chained together or started from a 360-degree panorama photo.

Google still calls both models frontier research. Its research overview says details and object permanence can drift over longer sessions or after repeated edits, and the announcement lists camera latency, complex physical interactions and real-time audio precision among unsolved problems. Generated video and audio carry SynthID, Google's invisible watermark for identifying AI-made content.

Key takeaways · AI-distilled
  • Genie 4 lets users type a new prompt mid-session to edit a world while moving through it, such as shifting the weather or adding a character, and for the first time pairs the visuals with real-time native audio that adapts to the scene, Google says.
  • Google kept Genie 3 in the picker for longer runs: sessions now last up to 3 minutes per generation, up from 60 seconds, multiple worlds can be chained into one journey, and new worlds can be seeded from 360-degree panorama images.
  • The Project Genie agent builds a world from a conversation by orchestrating Google's models for the initial visuals and a mood-matched background score; Gemini can also watch the screen and suggest destinations or trigger in-world events.
  • Google describes Genie 3 and Genie 4 as frontier research models with open problems in camera latency, intricate physical interactions, real-time audio precision and long-term scene permanence. All generated video and audio carries SynthID watermarks.
Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
Read the source blog.google
Recommended reads
Comments

Checking sign-in…

Loading comments…