By Olivia Hart

Google DeepMind, the AI lab within Alphabet's Google, has begun a limited rollout of Genie 4, a world model that generates real-time visuals and audio for interactive environments lasting up to two minutes. Google announced the model on Tuesday, and a limited group of trusted testers can now access it.

Genie 4 turns prompts into navigable scenes, extending Google's effort to build AI systems that understand how the world behaves. Google said simulating the world is a step toward artificial general intelligence, with the technology intended to support agents and creative work.

What can users control and hear?

Users can steer a scene with plain-language commands. They build each world by typing a prompt, then move through it with the keyboard, Bloomberg reported.

• Camera: Shift the camera angle.
• Weather: Alter weather conditions.
• Characters: Introduce new characters into the scene.

Genie 4 generates audio on the fly alongside the visuals. Its output can include:

• Ambient sound
• Music
• Interactive dialogue

The longer runtime does not guarantee a stable scene. Google DeepMind research engineer Hang Qi told Bloomberg: "Reaching the two-minute limit is no guarantee that a scene holds together from start to finish." TechCrunch reported that details and objects may shift or warp over time, particularly in longer simulations.

How does Google build each world?

Genie 4 builds on Google's Gemini family of models. Two components have distinct roles in each simulation:

• Gemini Omni: Anchors appearances and behavior in the real world.
• Gemini 3.8 Flash: Acts as a guiding agent, observing the scene and tracking progress as the world generates.

The arrangement gives each simulation a visual anchor and a separate task tracker. It also lets user commands alter a shared environment while the guiding model monitors what happens next.

Who gets access, and when?

Google has set a staged distribution plan for the model. The plan divides initial access between consumer subscribers and external partners.

• Google AI Ultra subscribers: Genie 4 will arrive soon through Project Genie.
• Early-access partners: Access will come through the Gemini API and AI Studio.
• Deployment window: The rollout is planned for the coming weeks, according to TechCrunch.

The distribution gives Google close feedback from trusted users before access widens. For practitioners, the next test is whether longer sessions deliver enough control to shorten early production work.

What changes for developers and filmmakers?

Trusted testers from game development and film are already using Genie 4 to turn ideas into prototypes that can be explored. They are also using it to rough out scenes, Bloomberg reported.

Google says the same technology could provide safe training environments for robots and agents. It also sees potential to add to creative tools in:

• Filmmaking
• Journalism
• Gaming
• Manufacturing

For game and film teams, the immediate use is exploratory, centered on testing ideas through prototypes and rough scenes. The disclosed limits prevent teams from treating the simulation as a finished production pipeline.

What remains unresolved?

Google DeepMind research director Tejas Kulkarni said fully consistent worlds across a two-minute stretch would require a model that grasps physics in daily life. According to TechCrunch, the team has not worked out how to ensure accurate audio or legible in-world text.

Diego Rivas of Google DeepMind told Bloomberg: "We're still in the research journey. We're not necessarily in the end product journey." Rivas described the project as early-stage, placing the initial release within continuing research rather than a finished service.

These gaps constrain a team's ability to judge continuity, dialogue and scene detail from a prototype. Unstable objects can undermine a concept, while inaccurate sound and unreadable text can limit testing.

How will Google identify generated output?

Google will tag every audio clip and video produced by Genie 4 with SynthID. Bloomberg reported that SynthID is the company's hidden marker designed to flag AI-generated material.

For users, the marker provides a provenance signal for generated clips. It addresses content identification, but not the separate problems of visual consistency, audio accuracy or text legibility.

What did the January debut reveal?

The January debut of Project Genie, a prototype built on Genie 3, rattled video game stocks, Bloomberg reported. The Genie 4 rollout adds longer sessions, live audio and plain-language scene controls.

Its disclosed limits leave developers with a research tool for testing ideas, not a guarantee that every generated world will remain coherent. That distinction will shape adoption as Google expands access beyond trusted testers.