ai-toolstutorialshow-toaigoogle

How to Prompt Project Genie and Prepare for Genie 4

Official tips to create interactive worlds in Project Genie today and what Genie 4 adds for real-time audio and scene steering.

October 28, 2026 Yum News

Project Genie lets eligible Google AI Ultra subscribers build and walk through short interactive worlds from text or images. Google DeepMind announced Genie 4 on 27 October 2026 as a research preview for trusted testers, with broader access planned for Ultra subscribers inside the same Project Genie experience. The newer model adds native audio and stronger real-time steering while you explore.

This guide shows how to create worlds with the current Project Genie tools, applies the official prompt tips, and covers what changes when Genie 4 arrives. For access status and waitlist details, see our earlier overview: /blog/how-to-access-genie-4-google-world-model/.

Who can use Project Genie today

Project Genie remains an experimental research prototype. Google rolled it out to Google AI Ultra subscribers in the United States who are 18 or older. It has expanded in stages and is listed as an Ultra feature in supported regions. You need a personal Google Account on an eligible plan.

If you do not yet have access, you can still open the public gallery and watch worlds other people have created. Generating new worlds does not consume your regular AI credits while the prototype is in this early stage.

Create your first world in four steps

Open labs.google/fx/projectgenie and sign in. Click Create with Project Genie. High demand can place you in a short queue; stay on the page until your turn arrives.

You then have 30 seconds to define the world. Choose one of three starting methods:

Write a text description of the environment and character.
Pick a Street View location and apply a style.
Use Roll the dice for a random starting point.

Select first-person or third-person view. You can tweak the generated preview image with Add, Change, or Remove tools. Click Next. Generation finishes in moments and exploration begins automatically.

Exploration lasts 60 seconds. Use standard keyboard controls:

W, A, S, D to move.
Spacebar to jump or ascend.
Arrow keys to turn the camera.

A progress bar at the top tracks remaining time. When the session ends you can review the result or start another world.

Person working at a computer with digital interfaces visible

Official prompt tips that improve results

Google published concrete guidance for Project Genie in March 2026. The same principles remain useful for the current experience and for preparing Genie 4 prompts.

Describe the environment in detail. Name the setting (forest, city, underwater, lunar surface) and add specifics: lush or wintry trees, busy or empty streets, cheese moon or realistic rock. Include structures, weather, and overall look—photorealistic, cartoon, or game-like.

Choose a character carefully. The character is your navigation point. It can be ordinary or fantastical. State how it moves (flies, rolls, hops) and any effects that follow its actions, such as smoke puffs or light trails.

Start from your own images when useful. Upload a photo where the character is centered and enough background is visible. Add text that modifies lighting, materials, or atmosphere. The preview uses Nano Banana Pro so you can refine before the full session starts.

Use short, direct language. “An underwater scene with dense coral, bright shafts of light, and a small goldfish character that swims upward when Space is pressed” works better than long paragraphs. Sensory details help set mood without overwhelming the model.

Switch perspectives after creation. First-person puts you inside the character; third-person shows the character from outside. The perspective button lets you change mid-session.

These steps come directly from Google’s prompt-writing post and the Project Genie launch materials.

What Genie 4 changes

Genie 4 is DeepMind’s newest world model. It builds on Gemini capabilities for appearance, behavior, and coherence. Key advances include:

Native audio generated on the fly: ambient sound, music, and interactive dialogue.
Real-time steering with natural language while the simulation runs. You can change weather, move the camera, or introduce new characters or objects without restarting.
Improved photorealism and fluid movement compared with earlier Genie versions.

Google has begun a limited research preview with trusted testers. The model is coming soon to Google AI Ultra subscribers inside Project Genie and to early-access partners through the Gemini API and Google AI Studio. A waitlist form is linked from the official DeepMind announcement.

Current Project Genie sessions stay short (setup plus 60 seconds of exploration). Genie 4 demos show continuous interaction, yet published materials do not yet list longer consumer session limits or exact pricing beyond the existing Ultra plan.

Watch the official announcement video for real-time examples:

Abstract digital technology and circuit patterns

Practical preparation steps

Confirm your Google AI Ultra subscription and region eligibility if you want the current Project Genie experience.
Practice the official prompt structure on any available session or by drafting prompts in Gemini first.
Join the Genie 4 waitlist from the DeepMind blog or models page so you receive notice when Ultra access expands.
Keep initial prompts concise. Add weather, character, or object changes as short follow-up instructions once Genie 4 steering is available.
Expect SynthID watermarks on generated outputs, consistent with other Google generative tools.

Limitations remain. Complex physics, long-horizon object permanence, and extended continuous sessions are still research areas. Treat Project Genie as an experimental creative tool rather than a production game engine.

Tips for better sessions

Center characters in reference images and leave clear background space.
Test one major change at a time when steering becomes available (weather, then new character, then camera).
Use first-person for immersion and third-person when you want to inspect the overall scene.
Review the public gallery for prompt patterns that produce stable results.
Check Google’s Project Genie help and the DeepMind Genie 4 post for the latest availability notes, since rollout timing can shift.

Conclusion

Project Genie already lets Ultra subscribers sketch short interactive worlds with text, images, or Street View and explore them for 60 seconds. The official tips—detailed environments, clear characters, direct language, and perspective switching—produce more coherent results today. Genie 4 will layer native audio and live natural-language steering on top of that foundation once it reaches Project Genie for Ultra users.

Start with the current prototype if you have access, draft prompts that follow the published guidance, and join the waitlist so you are ready when the newer model arrives. The combination of today’s controllable sessions and tomorrow’s audio-aware worlds opens new options for prototyping scenes in film, games, education, and research.