Google Makes Genie 4 Available in Project Genie to Some AI Ultra Subscribers

Google has begun gradually making the interactive world model Genie 4 available in the new Project Genie to eligible Google AI Ultra subscribers.

On November 18, 2026, Google announced the gradual rollout of Genie 4 Project Genie to eligible Google AI Ultra subscribers. The new version of the tool is intended to let users create interactive simulated environments and modify them with text prompts directly in the browser while navigating them.

Genie 4 belongs to the category of so-called world models. These are intended to connect the generation of visual content, sound, and interaction in real time. Google had previously presented this technology primarily as a research preview, but is now making it available to a selected group of paying users.

What Genie 4 Project Genie offers

According to Google, users can continuously enter text-based changes while moving through a generated world. For example, they can modify the weather, characters, or events in the environment. The model is also intended to add native audio, including dialogue.

The Project Genie update also includes the Project Genie agent. Google is additionally adding a community gallery where users are intended to share created worlds and remix the work of others. However, the announcement does not specify what specific capabilities the agent provides or the terms for using community creations.

Availability is currently limited to eligible Google AI Ultra subscribers. In the announcement, Google did not confirm general availability for all users. It also did not mention access to the model through the Gemini API or Google AI Studio, and did not clarify the scope of availability by country.

Watermarking and stated safety controls

Google says that video and audio created in Project Genie receive a SynthID watermark. This is the company’s technology for labeling content created by artificial intelligence. Inputs and outputs are also intended to undergo safety checks.

The announcement does not provide an independent assessment of the effectiveness of these measures or a detailed description of how they work in specific situations. However, Google’s model card provides an evaluation of the model and lists known system limitations.

Google acknowledges the model’s limitations

The company openly warns that results may not always be stable. The stated limitations include inconsistency in details and object persistence, meaning the ability to keep objects consistent within a scene. Difficult physics interactions, audio accuracy, and long-term scene consistency may also pose problems.

In practice, this means that a generated world may not preserve all elements as expected during longer interactions or more complex user interventions. Google has not published an independent comparison of Genie 4’s quality with competing world models.

What to watch next

The key questions in the near term will be whether Google expands Genie 4 availability beyond Google AI Ultra subscribers and publishes a list of supported countries. Possible access for developers through the Gemini API or Google AI Studio also remains an open question.

Independent tests of scene stability, physics interactions, responsiveness, and safety measures will provide more insight into the tool’s practical quality. The current announcement confirms limited consumer access through Project Genie, not general deployment of the model across Google’s products.

Sources

  • Google Blog – Confirms the announcement date, gradual availability to eligible Google AI Ultra subscribers, Project Genie features, SynthID, and stated safety checks.
  • Google DeepMind – Confirms Genie 4’s nature as an interactive world model, its native audio, availability in Project Genie, and technical limitations.
  • Google DeepMind – The model card describes the intended distribution through Project Genie, the model evaluation, and known limitations.

Verified and updated: November 19, 2026 06:25

Sharing