Google DeepMind unveils Genie 4 world model: Generate interactive 3D worlds in real time, with audio and character dialogue generated for the first time.
A world model doesn’t play a precomputed video. Instead, it calculates the next frame in real time, one frame at a time, based on your actions.
Key points:
Native audio: Music, ambient sounds, and character dialogue are generated alongside the visuals
Change scenes in real time: Type prompts during a simulation to change the weather, switch camera angles, or add people and objects
Consistency: Scenes can remain consistent for up to about 2 minutes—twice as long as with Genie 3
Gemini as director: Plans the story and checks frame by frame whether it’s following the plan
Robot simulation: Gemini 4 Argon breaks down tasks, allowing Genie 4 to simulate the humanoid robot Apollo playing chess, without using a physical robot
Background: When Project Genie (Genie 3) opened to the public this January, gaming stocks plunged in a single day. Unity fell as much as about 24%, Roblox about 13%, and Take-Two around 8% to 10%.
On the day of this announcement, Unity closed down 0.57%, while Roblox closed up 0.18%.
Availability: For now, only invited testers can use it. Over the next few weeks, access will open to Google AI Ultra subscribers and early partners using the Gemini API and AI Studio. The public release date and API pricing have not yet been announced.
Limitations: Over time, visuals can “drift”; sessions last only a few minutes, audio isn’t very accurate, and text in images can be hard to read. All content will include a SynthID watermark.
A world model doesn’t play a precomputed video. Instead, it calculates the next frame in real time, one frame at a time, based on your actions.
Key points:
Native audio: Music, ambient sounds, and character dialogue are generated alongside the visuals
Change scenes in real time: Type prompts during a simulation to change the weather, switch camera angles, or add people and objects
Consistency: Scenes can remain consistent for up to about 2 minutes—twice as long as with Genie 3
Gemini as director: Plans the story and checks frame by frame whether it’s following the plan
Robot simulation: Gemini 4 Argon breaks down tasks, allowing Genie 4 to simulate the humanoid robot Apollo playing chess, without using a physical robot
Background: When Project Genie (Genie 3) opened to the public this January, gaming stocks plunged in a single day. Unity fell as much as about 24%, Roblox about 13%, and Take-Two around 8% to 10%.
On the day of this announcement, Unity closed down 0.57%, while Roblox closed up 0.18%.
Availability: For now, only invited testers can use it. Over the next few weeks, access will open to Google AI Ultra subscribers and early partners using the Gemini API and AI Studio. The public release date and API pricing have not yet been announced.
Limitations: Over time, visuals can “drift”; sessions last only a few minutes, audio isn’t very accurate, and text in images can be hard to read. All content will include a SynthID watermark.
