Hey there,

Back in August 2025, Google shocked the AI world with their new research model that could turn simple images into completely new, interactive worlds. The Genie 3 demos were incredibly impressive and showed us yet another application of AI in the near future. 

Now that future is here.

Yesterday Google released Project Genie, a new experimental prototype that turns simple prompts and images into fully explorable worlds. These AI worlds are playable, interactive environments that can be anything from conducting a spacewalk to skydiving through the mountains. So, how do you build these worlds, and what can they do? Let’s take a look. 

From a Prompt to an Entire World

Project Genie is a web app powered by Google’s most powerful models: Gemini 3, Nano Banana Pro, and Genie 3. This new experience is centered around three core creative paths.

World Sketching: 

When creating a world in Project Genie, you can use the new “world sketching” tool to see a preview of your creation. By using Nano Banana Pro, you can quickly see a preview of what your prompt will generate and make any necessary tweaks without needing to fully generate a project. In the example below, I changed the weather of my world multiple times after creating a sketch. 

World Exploration: 

Building your world is just the beginning, now you can actually explore it. As you move through your creation, Genie is actively generating the path ahead in real time. Each world is a unique live experience, not a pre-rendered level, so your actions shape what is created. 

World Remixing: 

Like so many other AI tools, iteration is essential. Once you’ve created your initial world, you can now refine the experience to better fit your expectations. This effectively means you can change your character, environment, or background while keeping the elements that work. You can also choose to remix one of the template worlds that Google has made public.

Managing Expectations: 

While Project Genie is incredibly exciting, it's important to remember that this is still the early stages for world models. There are some significant limitations to keep in mind:

• Fidelity: Worlds aren’t always “realistic”. Physics can often act weird or you may clip through objects. 

• Control: Characters can often feel hard to control or sometimes completely unresponsive. They can also be laggy depending on the complexity of the world. 

• Length: Generated experiences are currently limited to just 60 seconds. This limit exists for a few reasons but most importantly because the world's quality starts to decrease as the experience goes on. 

How to Access It

Due to its experimental status, Project Genie has heavily restricted access. Only Google AI Ultra subscribers in the U.S. can access the tool. Limiting access to Ultra users means that they can slowly roll out the experience without needing to worry about overwhelming their systems.

My Thoughts:

To understand why Project Genie is so impressive, you first need to understand the technology behind it. It’s powered by what’s called a “world model”. Unlike a traditional LLM that predicts the next word, a world model tries to predict and simulate the future of an entire environment. This allows it to create a dynamic space where the user’s actions impact the world. It's a foundational step toward AI models better understanding complex topics as well as cause and effect. That’s what makes Project Genie one of the most impressive AI tools of 2026. This is more than just a “cool gaming tool”; it's a major step toward building world models that can understand and simulate complex, dynamic environments. Whether you’re prototyping a new game or designing an interactive learning experience, having a tool like Genie 3 allows you to experiment with AI in an entirely new way. 

If you want to experiment for yourselves or just check out some interesting examples, head to Project Genie to get started.