Google Genie 4: how to get access and what the world model can do

How to access Genie 4

Google’s announcement offers a waitlist for Genie 4 in Project Genie. Joining expresses interest; it does not guarantee access or a particular invitation date.

1. Open the waitlist linked from Google DeepMind’s announcement.
2. Review and complete Google’s form according to its instructions.
3. Watch for Google’s access information. Verify eligibility and current terms before buying a subscription.

Project Genie is an application; Genie 4 is the underlying world model. Project Genie previously launched as a Genie 3-powered prototype for eligible U.S. Google AI Ultra subscribers aged 18 or older. That earlier rollout does not establish Genie 4’s final country or eligibility rules. See Google’s Project Genie announcement.

Question	What is confirmed as of October 28, 2026
Who has access?	Trusted testers are receiving a research preview. Google plans subsequent access for Ultra subscribers in Project Genie and selected developer partners.
Release date and countries	Google’s Genie 4 announcement does not specify a general release date or a complete country list.
Price	Google has not specified a separate Genie 4 price or public API rates in its launch announcement. Ultra subscription terms are separate and can change.
API and model ID	Early access partners are planned for the Gemini API and AI Studio. The launch announcement does not provide a public Genie 4 model ID or integration instructions.

Do not assume an existing Ultra subscription or Gemini API key already grants Genie 4 access. Check your account and Google’s current documentation.

What can Genie 4 do?

Google’s launch demonstrations show recorded real-time interactions. They illustrate research capabilities, not a promise that every feature is available to every user.

• Generate sights and sounds together. Google says Genie 4 can generate music, background sounds and conversations as the world unfolds.
• Navigate and steer a scene. Demonstrations use keyboard controls for navigation and camera movement, with text instructions to change a scene as it evolves. Google describes Gemini adapting those instructions to the surrounding context.
• Simulate people and events. The demonstrations show people moving, reacting and speaking, as well as scenes with multiple steps. These are generated simulations, not recordings or verified predictions of real events.
• Follow multi-step events. In Google’s examples, Gemini 3.8 Flash helps plan and guide actions as the scene changes, while Genie generates what you see and hear.

Watch the official Genie 4 demonstration and turn on audio to assess the examples yourself. Google’s article also includes the prompts used for its examples.

How to judge the demos

In Google’s lake scene, a musician plays on a floating platform while the camera moves independently. Watch whether the musician and platform remain consistent as the viewpoint changes. In the complex-event examples, watch whether each action follows the previous one; in the audio examples, listen for sound matching the visible event. Google provides the example prompts alongside the clips.

These are selected recordings, and Google says clips have been shortened. A clip’s length does not establish how long a session ran, and a polished example does not establish the success rate for other prompts.

A prompt to prepare for access

Use the structure in Google’s examples to plan a scene: describe the environment, subject, viewpoint, motion and sound. The example below is a starting point; we have not tested it in Genie 4, and the available controls may differ.

First-person view on a quiet coastal path at dusk. Walk slowly toward a small lighthouse while waves break on the rocks. Keep the lighthouse visible as the camera turns, with wind and distant seabirds in the background. Change the weather to light rain as I approach.

Once invited, try one change at a time. Turn away and return to check whether the lighthouse remains in place; compare the sound with the visible weather; and note what changes after the rain instruction.

What is a world model?

A world model simulates aspects of an environment and how it changes in response to actions. In Genie’s case, the experience generates the view as a person navigates or changes a scene. Unlike a conventional finished video, the sequence can respond to interaction, though that does not make its physics or facts reliably accurate.

Genie 4 vs an AI video generator: what is the difference?

The practical difference is that you can explore and change a Genie 4 scene as it unfolds. A typical generated video plays a finished clip. In Google’s Genie 4 demonstrations, a person navigates while the scene is generated and can use text to steer what happens next. Both can produce convincing visuals and audio, but neither should be assumed to model real physics perfectly.

Question	Typical generated video	Genie 4 research demonstrations
How do you direct it?	Describe a clip before generation; some tools support later edits.	Navigate with keyboard and mouse, and use text to steer the evolving scene.
What do you get?	A clip to watch or edit.	An interactive, generated visual and audio experience in Google’s demonstrations.
What should you verify?	Whether the clip meets your editing and licensing needs.	Whether controls, consistency, duration and available exports meet your task. Google has not confirmed a downloadable game world or asset export.

Some video tools also offer editing or interactive features. Google has not confirmed that Genie 4 creates a saved 3D world or a project you can open in a game engine.

Genie 4 vs Genie 3: what changed?

Both let you explore generated worlds. Here is what Google has said about each model; the differences have not been independently tested here.

Area	Genie 3	Genie 4
Interaction	Real-time navigation and promptable world events were already demonstrated.	Google demonstrates keyboard navigation and natural-language changes while the scene evolves, with Gemini helping steer it.
Sound	Google’s original announcement focused on visual world generation.	Google describes native audio generated with the world, including ambient sound, music and dialogue.
Multi-step events	Google demonstrated longer sequences and agent research.	Google demonstrates a Gemini agent observing and guiding multi-step simulations.
Published performance	At its 2025 announcement, Google reported 720p, 24 fps and consistency for a few minutes.	Google’s model page describes a few minutes of continuous interaction, without giving a precise limit; the launch announcement gives no comparable resolution or frame rate.
Access	Project Genie launched as a Genie 3-powered prototype for eligible U.S. Ultra subscribers.	Research preview for trusted testers, with further staged access planned.

Sources: Genie 3 announcement, Genie 4 announcement and Project Genie launch. Genie 3’s resolution, frame rate and Project Genie session limit do not tell us what Genie 4 will offer. Google’s Genie 4 model page describes a few minutes of continuous interaction, but does not give an exact limit for the upcoming Project Genie experience.

What are Genie 4’s limitations?

Google’s model page says continuous interaction currently lasts a few minutes, rather than extended sessions. It does not specify an exact maximum. Objects and visual details can change or disappear, particularly during longer sessions, after several edits or as you move around. Complex physics, accurate real-time audio, readable text, spatial accuracy and long-term consistency remain research challenges. Google also identifies multi-agent interaction and accurate reproduction of real locations as limitations. A convincing demo is not proof that a simulation is physically accurate or suitable for consequential real-world decisions.

Google also says generated video and audio will carry SynthID watermarks. Its launch post describes internal safety evaluations and an initial trusted-tester rollout.

What might people use it for?

Google identifies interactive experiences and training environments for robots and agents as potential applications. A creator could explore a scene concept; a researcher could investigate a simulation workflow. These are possibilities described by Google, not evidence that Genie 4 is ready for production games, robot deployment or safety-critical simulation.

A creative workflow shown by a Genie 4 contributor

In a public demonstration, Genie 4 contributor Tejas Kulkarni describes starting with a Nano Banana image, iterating on animations in Genie 4, then using the result as a reference for Argon to build and sculpt in Blender. The example uses several tools: it does not show Genie 4 directly exporting an editable 3D model, and the same process may not be available to everyone.

What to test when you get access

When you get access, try a scene that matters to you. These checks can help you decide whether the results are useful for your project.

1. Does the world stay consistent? Put a distinctive object in view, turn away, then return. Check whether it still looks the same and is in the same place. Try again after changing the scene.
2. Does it respond to you? Move around, then request one change in text. Check whether it happens and whether other parts of the scene change unexpectedly.
3. Do sound and people make sense? Listen for sounds that match the action and conversations that follow the scene. Try the same idea more than once to see how the results vary.
4. Can you use the result? Time your session and check which saving, sharing and export options are available in your account. Google has not published a specific session limit for Genie 4 in Project Genie.

A short scene may help you explore a film or game idea. For a finished game or a robotics project, you would also need to check how reliably the world behaves. Google has not confirmed that you can export game assets, save a persistent world, use multiplayer or access Genie 4 in VR.

Genie 4 FAQ

Sources

This guide was checked against Google’s announcements on October 28, 2026. We have not tested Genie 4 ourselves. Access, pricing and capabilities may change.

• Google DeepMind: Genie 4 announcement — capabilities, rollout, limitations and waitlist.
• Google DeepMind: Genie 4 model page — interaction duration, edited demonstrations and limitations.
• Google DeepMind: Genie 3 announcement — prior capabilities and published specifications.
• Google: Project Genie launch — prototype and initial eligibility.
• Google DeepMind: Genie 4 demonstration — official video examples.

Related guides

Gemini 4 Argon

Explore a separate Gemini model and its announced access status.

Run Gemma 4 locally

Learn about a different model family designed for local use.