How to Create an Interactive World with Project Genie

Project Genie is a Google Labs research prototype powered by the Genie 3 world model. It generates an interactive environment as you move through it, rather than producing only a fixed image or video. You can describe a world with text, use an image as a starting point, choose a character and camera view, then remix the result.

Project Genie is gradually available to eligible Google AI Ultra subscribers aged 18 or older. It is an experiment, not a conventional game engine: visual details, controls, object permanence, and physical behavior may be inconsistent. Check the live product page for account and regional access instead of relying on older U.S.-only availability information.

1. Describe a world with navigable structure

Begin with the setting, but include features that give the model somewhere to go. A prompt such as “snowy forest” establishes a look; adding a path, clearing, bridge, distant landmark, and weather creates a more useful space.

A stylized third-person adventure world in a snowy pine forest at dawn. A narrow trail curves past frozen ponds toward a warm cabin on a ridge. Light snowfall, soft blue shadows, and clear space around the trail for movement.

Define only the most important visual style and lighting. Too many conflicting labels—photorealistic, pixel art, watercolor, and cinematic at once—make the direction harder to follow.

2. Design the character around movement

The playable character can be realistic or imaginary, but its shape and locomotion should suit the environment. Describe:

• its scale relative to the world;
• how it moves, such as walking, rolling, flying, or swimming;
• one recognizable color or silhouette;
• any restrained movement effect, such as footprints or a short light trail.

Explorer: a small orange rover with four wide wheels and a white antenna. It drives steadily over rough ground and leaves faint tire tracks in snow.

Start with one character and one movement mode. Add unusual behavior only after the basic world is controllable.

3. Use a reference image with room to explore

A source image can anchor the composition, colors, or subject. Choose a clear image with the main subject separated from the background and enough surrounding context for the model to infer an environment. A tightly cropped portrait offers little information about terrain or direction.

After upload, explain what to preserve and what may change:

Keep the red lighthouse, rocky shoreline, and overcast palette from this image. Extend the coast into a walkable path, add shallow tide pools, and keep the lighthouse visible as the navigation landmark.

Do not upload confidential material or images you lack permission to reuse. For related image-based workflows, TipsMake explains how AI tools turn 2D images into 3D assets.

4. Prompt in layers and revise one problem at a time

Use direct sentences that separate environment, character, movement, camera, and mood. Generate a preview, inspect it, then change the weakest element. If the path is unclear, revise the terrain; if the character is hard to see, change its scale or contrast; if motion feels confusing, simplify the scene.

A practical prompt template is:

World: [setting and navigable landmarks]
Style: [one visual style, lighting, and weather]
Character: [appearance and scale]
Movement: [how the character moves]
Camera: [first-person or third-person]
Priority: [the one detail that must remain consistent]

Project Genie can show environment and character previews before exploration. Treat those previews as drafts, not guarantees of every frame. If you want to combine visual references for a separate video workflow, see the guide to combining reference images in Google Flow.

5. Choose first-person or third-person deliberately

First-person places the camera at the explorer's viewpoint. It suits architectural interiors, walking simulations, and environments where scale is important. Third-person keeps the character visible, making it easier to judge movement and character effects.

Use the view switch in the exploration interface to compare both. If a scene was designed around a visible character, the first-person version may lose that visual anchor; if a narrow interior is difficult to navigate in third person, the closer view may work better.

Evaluate the world before remixing it

• Can you identify the route or landmark immediately?
• Does the character remain recognizable while moving?
• Does the camera suit the scale of the scene?
• Do important objects persist well enough during a short exploration?
• Which single prompt change would improve the next version?

Save or share only after checking the current result. Because the world is generated dynamically, another run or remix may not reproduce every visual detail.

What Project Genie is—and is not

Project Genie is useful for exploring world concepts and studying interactive generative environments. It does not replace a production game engine, level editor, physics simulator, or verified digital twin. Do not use its output to make safety, engineering, navigation, or accessibility claims about a real place.

Google's Project Genie announcement describes the prototype and its research purpose. The Genie 3 model page provides the current overview and access route.