Google Genie 4 vs Genie 3: What changed in Google's world model

On November 18, Google released an update to its Genie world model while keeping the previous version intact. The latest version of Genie called Genie 4 is currently live on Project Genie in case of those eligible for Google AI Ultra subscription, featuring improved visuals, physics, and audio in real time. The previous version of Genie named Genie 3 remains accessible from the model selector but has been updated with its own improvements. Thus, there are two world models available in the same browser tab, and you have to decide which one to use. But what changes did Google make, what is still better with Genie 3, and what other updates were made?

What does Genie 4 add over Genie 3

The Genie 4 by Google DeepMind is its most realistic and interactive world model till date, with features such as an ability to modify the world you are standing in while playing it out. You can type a new prompt during a session and the world will change accordingly. It could include changing the weather, introducing a new character, or simply redirecting the scenario without restarting from scratch.

The other notable feature in this model is native audio, which means that Google DeepMind has added sound to this model, which reacts according to the evolving scenario. As per Google DeepMind, the Genie 4 gives its users the best visual and audio experience.

Does Genie 3 still matter

Yes, and Google keeps this feature in the model picker for sessions that last longer and panoramas for seeding. Also, it got three upgrades. Session lengths have been increased to three minutes per generation, from one minute, and users can combine several worlds into a long journey. A new world can be generated based on a 360-degree panorama image. In addition, real-time events in the world will allow you to make changes in the environment using natural language prompts.

This is important since the ability to edit worlds mid-session already exists in Genie 3. Therefore, the difference between Genie 3 and Genie 4 is mostly visual, and not in terms of the ability to make changes. There is no limit on sessions for Genie 4 in the announcement.

What is the Project Genie agent

The agent turns world-building into a conversation rather than a text or image prompt. Explain your world in simple terms, and the agent uses Google’s multimodal models to create the first visuals and generate an ambient soundtrack that matches the mood.

Gemini models can participate in the experience. They observe the action on the screen, show suggestions as to what comes next, and set off in-world events as you explore. Think of a co-pilot who is also a game master.

The gallery blends worlds made through curation with those made through community efforts, and both may be remixed at will. Modify the structure, audio, or game rules, then publish the remix. Living Memory transforms your own pictures into interactive worlds, and Rock Paper Scissors features Gemini mediating a battle of items on a desk.

Two Genie 4 exhibits were created by external collaborators. Villa Giallo, created with Will aka BLVCKL!GHT, is an unfolding mystery throughout a gothic mansion. Life of Ants, created with filmmaker Anand Gandhi, places you within the nest and lets you experience the world from an ant’s point of view.

Who can use it and what are the limits

Qualified members of Google AI Ultra will be able to gain access from the browser starting from November 18, while those in the Google AI Pro program will be able to sign up for the waiting list. No regions have been specified in the announcement, so readers in India can visit the Project Genie website.

Both of the mentioned systems are described by Google as being on the cutting edge of research. Camera lag, complex physics, real-time audio, and scene persistence are among the unsolved issues. Watermarking with SynthID is applied to all video and audio, and all inputs and outputs undergo safety checks.

Vyom Ramani

Vyom Ramani A journalist with a soft spot for tech, games, and things that go beep. While waiting for a delayed metro or rebooting his brain, you’ll find him solving Rubik’s Cubes, bingeing F1, or hunting for the next great snack.

View full profile