Remember Google Genie 4?
Google's AI world model that generates interactive environments in real time?
Well, it's now available through the Gemini API and Google AI Studio.
So naturally, I decided to test it by asking for GTA 6.
And the result is in this video.
I got a GTA-inspired world that I could actually explore, complete with characters, streets, and even a billboard featuring my own "Dots on AI" branding.
And here's the part I found especially interesting.
I could talk to the NPCs.
Unfortunately, my microphone wasn't captured in the recording, so you can't hear my side of those conversations. But I was interacting with the characters by voice while exploring the environment.
WHAT EXACTLY IS GENIE 4?
Genie 4 is a real-time world generation model developed by Google DeepMind.
Unlike traditional AI video generators, it doesn't simply produce a fixed sequence of frames.
It generates an environment that responds to your actions.
You can move through it, turn the camera, interact with characters, and change what's happening using natural language.
The world is continuously generated as you explore.
WHAT CAN IT DO?
• Create interactive worlds from text prompts, reference images, or both.
• Generate video at 720p and 24 FPS.
• Produce synchronized audio, including environmental sounds, music, and character dialogue.
• Let you navigate using WASD and camera controls.
• Change weather, lighting, objects, and events while exploring.
• Introduce new characters and direct their actions.
• Generate dialogue that reacts to events and user input.
• Connect world generation to custom applications through the Gemini API.
This last point is particularly important.
You can now experiment with AI-generated interactive environments inside your own software instead of being limited to Google's Project Genie interface.
MY GTA EXPERIMENT
For this test, I asked Genie 4 to generate a GTA-inspired environment.
And yes, it accepted the request.
The result wasn't an actual copy of Rockstar's game or a functioning version of GTA VI.
It was a dynamically generated world based on my prompt.
The model even included a billboard with my own show's branding.
I also tested voice interaction with NPCs.
My voice wasn't included in the screen recording, but the interaction itself was part of the experiment.
That combination of real-time world generation, navigation, and conversations is what makes Genie 4 so interesting to me.
WHAT ABOUT COPYRIGHT?
One interesting observation from my experiment is that the model accepted a prompt referencing GTA 6.
However, accepting a prompt doesn't mean the generated content is automatically free of copyright or trademark restrictions.
Google's API terms still require users to respect third-party intellectual property rights.
So it's better to think of this as a GTA-inspired AI experiment, not an officially licensed Grand Theft Auto experience.
THE TECHNICAL LIMITATIONS
Genie 4 is still an experimental world model.
Current API limitations include:
• 720p video resolution.
• 24 frames per second.
• A maximum of 120 seconds per API session.
• Noticeable delay before the first frame appears.
• Input latency during movement.
• Visual inconsistencies and hallucinations.
• Objects and environmental details that may drift over time.
The last problem is especially important.
In a traditional game engine, a building has a defined position and geometry.
In Genie 4, the model is generating its representation of the environment continuously.
That makes long-term object permanence, consistent geography, and reliable interactions difficult.
After extended exploration, the world may gradually become less consistent.
And this is where the technology still has a long way to go.
THE PRICE OF GENERATING A WORLD
Genie 4 isn't available on the free API tier.
The effective price of generated video is approximately $0.10 per second.
That means:
• 10 seconds: about $1.
• 1 minute: about $6.
• 2 minutes: about $12.
• 1 hour: about $360, hypothetically.
The hourly figure is just a cost extrapolation. You can't currently run one uninterrupted hour-long session, because individual sessions are limited to two minutes.
Additional input and text-output charges may also apply.
So yes, you can generate your own interactive world.
But spending an evening exploring it would be a rather expensive hobby.
PROJECT LINKS
Google AI Studio:
https://aistudio.google.com/
Official Genie 4 API announcement:
https://blog.google/innovation-and-ai/technology/developers-tools/genie-4-in-gemini-api/
Genie 4 research:
https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
Project Genie:
https://labs.google/projectgenie
Google AI Studio Cloud credits:
https://blog.google/innovation-and-ai/technology/developers-tools/google-cloud-credits-gdp-ai-studio/
#Genie4 #GoogleDeepMind #AIWorldModels
YouTube / ABVX
