Google just opened Genie 4 to developers through the Gemini API and Google AI Studio. And this is a significant step for generative world models. Instead of generating a video that you simply watch, Genie 4 creates an interactive environment that responds to your movements and text instructions in real time. You can walk around, turn the camera, change the weather, modify objects, and trigger events inside the generated world. All with synchronized AI-generated audio. But there's a catch. It costs approximately $6 per minute, and a session can last no longer than 120 seconds. WHAT CAN GENIE 4 DO? • Generate an interactive world from a text prompt, a reference image, or both. • Stream continuous 720p video at 24 FPS. • Generate synchronized native audio, including ambient sounds, effects, and character dialogue. • Navigate using WASD controls. • Rotate the camera with the arrow keys. • Change weather, lighting, objects, and events using text prompts during the session. • Trigger character dialogue and other world events. • Integrate interactive world generation into custom applications using the Gemini API. The model doesn't just render a predetermined video. It generates the next frames based on what the user does. HOW RESPONSIVE IS IT? According to Google's documentation: • Initial world generation takes approximately 5–9 seconds before the first frame appears. • Keyboard movement responds visually in approximately 0.2–0.5 seconds under favorable network conditions. • Mid-session text instructions generally take 1–2 seconds to affect the environment. • Video output is fixed at 1280 × 720, 24 FPS, in a 16:9 aspect ratio. These are documented expectations, not guarantees for every user or network connection. HOW MUCH DOES IT COST? Genie 4 has no free tier in the Gemini API. Google charges $17.50 per million output video tokens. At 24 FPS, that works out to approximately: • $0.10 per second. • $6.05 per minute. • $12.10 for a full 120-second session. Synchronized audio is included in the video rate. Input tokens are billed separately. And here's an important warning for developers: closing a browser tab doesn't necessarily stop the server-side generation. Your application must explicitly complete the session to stop the billing. THE BIGGEST LIMITATION: WORLD MEMORY Genie 4 doesn't maintain a traditional persistent 3D scene graph. It generates visual frames using a rolling temporal context. That means the world can gradually change as you explore it. Turn around, look away from an object for several seconds, and then return. The object might have moved. A doorway might look different. A building might not be exactly where it was before. Google warns that visual consistency can degrade after approximately 60–90 seconds. And 120 seconds is the hard session limit. So no, this isn't a replacement for Unreal Engine, Unity, or a conventional game world. Not yet. THE DEMO: VILLA GIALLO The video shows Villa Giallo, an interactive mystery experience created by Google in collaboration with creative technologist Will, also known as BLVCKL!GHT. It's a gothic mansion you can explore room by room. Instead of watching a prerecorded walkthrough, the world is generated as you navigate the environment. Villa Giallo was showcased as part of the Project Genie launch, demonstrating the creative possibilities of Genie 4. NOW DEVELOPERS CAN EXPERIMENT TOO Previously, Genie 4 was primarily accessible through the new Project Genie experience for eligible Google AI Ultra subscribers. Now developers with a paid Gemini API key can build their own interactive world experiences. Google provides examples for JavaScript, TypeScript, Python, WebRTC, and WebSocket integration. For browser applications, the recommended architecture is to create the session on your backend and give the browser a short-lived, session-specific authentication token. Never expose your permanent Gemini API key in client-side code. TRY GENIE 4 Google AI Studio: https://aistudio.google.com/ Genie 4 API documentation: https://aistudio.google.com/docs/genie Official announcement: https://blog.google/innovation-and-ai/technology/developers-tools/genie-4-in-gemini-api/ Project Genie: https://labs.google/projectgenie Google's Villa Giallo showcase: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/genie-4-now-available-project-genie/ Would you pay $6 per minute to explore an AI-generated world? Or does this technology need to become dramatically cheaper before it becomes useful for real games? #Genie4 #GoogleDeepMind #AIWorldModels #GeminiAPI #Shorts