Genie 4 is set to become available to Google AI Ultra subscribers
Genie 4 is expected to become available to Google AI Ultra subscribers through Project Genie in the coming weeks. For now, selected testers are trying it. The company highlights limitations involving longer simulations, complex physics and audio accuracy.
According to Google, Genie 4 is expected to become available in the coming weeks to Google AI Ultra subscribers through Project Genie and to partners with early access to Gemini API and Google AI Studio. It is currently available only to selected testers in research preview. A waiting list is open for those interested in trying it through Project Genie.
Google DeepMind introduced a model for generating interactive environments frame by frame in real time. Genie 4 adds music, ambient sounds and interactive dialogue. During a simulation, users can use text prompts to change the weather or camera angle, or add characters and objects. According to the company, the model can maintain a consistent scene for up to two minutes, compared with approximately one minute for Genie 3, and improves both the depiction of people and physics simulation.
According to the company, Gemini 3.8 Flash continuously monitors the scene, breaks goals down into steps and checks new frames to verify whether those steps have been completed. When conditions change, it adjusts the plan. In a research demonstration, Gemini 4 Argon broke a game of checkers down into steps that Genie 4 simulated with a virtual robot; no real robot took part in the experiment.
The company also acknowledges that details and objects may lose consistency during longer simulations or repeated complex edits. Complex physical interactions, accurate real-time audio and legible text remain subjects of research. According to the company, all generated videos and audio tracks are intended to carry an invisible SynthID watermark.
Why it matters
Ongoing text-based edits allow users to try different versions of the same scene while it runs. The company cites potential uses in preparing film scenes, testing game ideas and creating robotic scenarios. However, this is an experimental approach with limited availability; its practical benefits for these tasks have not yet been independently verified.
Release card
Genie 4
Google DeepMind
- Inputs
- The model accepts text instructions for modifying the simulation and creates an interactive visual environment with sound, including music, ambient sounds and character dialogue.
- Availability
- The model is available to selected testers in research preview. In the coming weeks, it is expected to become available to Google AI Ultra subscribers through Project Genie and to partners with early access to Gemini API and Google AI Studio.
- It allows users to change the weather, camera angle or objects in an interactive scene as it runs.
- According to the company, it can be used to prepare film scenes and storyboards.
- According to the company, it can be used to prototype game ideas.
- During longer simulations or repeated complex edits, details and objects may lose consistency.
- Complex physical interactions remain a subject of research.
- Accurate real-time audio and legible text are not yet reliably supported.
According to Google DeepMind, compared with Genie 3, the model adds audio generation and ongoing environment edits through text instructions, improves the depiction of people and physics simulation, and extends scene consistency from approximately one minute to up to two minutes.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
What was added since the original report
Verified updates
-
Access is expected to expand to Google AI Ultra subscribers later.; According to the company, the model has limitations involving longer simulations and complex physics.
- Access is expected to expand to Google AI Ultra subscribers later.
- According to the company, the model has limitations involving longer simulations and complex physics.
-
The generated audio includes music, ambient sounds and interactive dialogue.; The Gemini model plans steps and checks frames while working with Genie 4.
- The generated audio includes music, ambient sounds and interactive dialogue.
- The Gemini model plans steps and checks frames while working with Genie 4.
-
Google reports improved physics simulation compared with Genie 3.; Google reports improved simulation of people compared with Genie 3.; The model generates environments frame by frame in real time.
- Google reports improved physics simulation compared with Genie 3.
- Google reports improved simulation of people compared with Genie 3.
- The model generates environments frame by frame in real time.
Two audiences, two different impacts
What this means
For individuals
A Google AI Ultra subscription alone does not yet guarantee access to Genie 4. Individuals who want to try creating their own interactive scenes can request access through the waiting list.
For a business
For companies considering robotic simulations, unresolved issues with complex physical interactions pose a risk to the fidelity of test scenarios. The sources describe a research demonstration, not a validated deployment for testing real robots.
DevelopmentCheck the original
Event sources
confirmed by 3 independent sources · 4 publishers, 3 independent. We count feeds from the same owner only once.