Skip to content
Theme
OnDevice Tools

AIGenerative AI

Genie 4 arrives in Project Genie with real-time audio and editing

Google DeepMind has released Genie 4 in Project Genie for eligible Google AI Ultra subscribers. The update adds mid-session world editing, native audio, a creation agent and a community gallery.

Genie 4 world model shown in a browser with real-time audio and editing controls

Key takeaways

  • Genie 4 is rolling out to eligible Google AI Ultra subscribers in Project Genie.
  • The model allows text prompts to change a world during a session.
  • Genie 3 now supports three-minute generations, panorama seeding and linked worlds.
  • Project Genie adds an agent, a gallery and SynthID watermarking.
In this article
  1. Genie 4 reaches Project Genie
  2. Real-time editing and native audio
  3. Genie 3 gains longer sessions
  4. Project Genie agent and gallery
  5. Safety, limits and availability

Genie 4 reaches Project Genie

Google DeepMind and Google announced on 18 November 2026 that Genie 4 has arrived in Project Genie. The rollout is for Google AI Ultra subscribers who are eligible. The information comes from official announcements.

Project Genie began earlier this year as an experimental research prototype. It lets people create and explore interactive worlds from a browser. A world model is an AI system that creates interactive environments. Google says early testers wanted deeper control and easier ways to build and share worlds.

Real-time editing and native audio

Google says Genie 4 can change a scene while a person explores it. A user can type a new prompt during a session. This can alter the environment. Examples include changing the weather, adding a character, or shifting the action.

The model also adds native audio. This is audio produced with the scene. It responds as the scene develops. Google says it covers background sound and spoken character lines.

The announcement says this is the first time Genie pairs visuals with real-time native audio. Google describes Genie 4 as its most realistic and interactive world model yet. It says the model offers improved visuals, physics and audio. Google presents these as improvements over earlier models.

Genie 3 gains longer sessions

Genie 3 remains available in the Project Genie model picker. Google says it is useful for longer runs. It is also useful for worlds seeded from panorama images.

The update extends each generation to up to three minutes, up from 60 seconds. It also allows multiple worlds to be linked into a longer journey. Another addition is 360-degree panorama image grounding. This can seed a new world from a panorama image.

Google also says Genie 3 supports real-time in-world events. These let a person change the environment with natural-language prompts. The company shows examples such as forests, underwater ruins and icy seas. It says panorama grounding can help keep landmarks consistent as the view changes.

Illustration: AI, Google logo, Gemini logo, privacy and the labels Gemini, Genie 4, Google AI Ultra

The new Project Genie agent is designed to make world building more conversational. A person can describe a vision in natural language. Google says the agent coordinates its multimodal models, which can handle different kinds of input. They create initial visuals and a background score.

Gemini models can also join a session. According to Google, they can observe the screen. They can offer context-aware suggestions for where to go next. They can also trigger in-world events during the journey.

A new gallery collects curated and community-built worlds. It includes Living Memory, Dino Field Guide and Rock Paper Scissors. Living Memory turns personal photos into explorable spaces. Dino Field Guide uses panorama-grounded prehistoric habitats, while Rock Paper Scissors has Gemini acting as a referee.

Google also highlights collaborations with creators. Villa Giallo is a room-by-room mystery made with creative technologist Will. The Life of Ants is a colony exploration made with filmmaker Anand Gandhi.

Users can publish creations to the gallery for others to explore and remix. The agent can also remix existing gallery experiences. It lets users adjust world structure, soundtrack or rules.

Safety, limits and availability

Google says all video and audio generated in Project Genie includes SynthID watermarking. This marks generated media. Google also says model inputs and outputs go through safety screenings. These measures are part of its responsible development approach.

The company describes Genie 3 and Genie 4 as world models in the frontier research space. It says work remains on camera latency. It also cites complex physical interactions, real-time audio precision and long-term scene permanence. These limitations mean the experience is experimental rather than a finished consumer product.

Access is limited to Google AI Ultra subscribers who are eligible. Google points readers to its Genie 4 model research overview. That overview covers limitations and progress. The official announcements provide current availability details.

Sources

  1. Genie 4 is now availableGoogle Gemini
  2. Genie 4 is now availableGoogle DeepMind

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.