Last updated: 2026-07-27
Project Genie is Google DeepMind's web-based world-generation prototype, launched January 29, 2026 on Google Labs. Built on Genie 3, an 11-billion-parameter transformer, it turns a prompt, image, or sketch into an explorable 3D world rendered at 24 frames per second, letting users walk, fly, or drive through AI-generated environments in real time.
About Project Genie
Project Genie is an experimental research prototype from Google DeepMind that generates interactive virtual environments from a prompt, an image, or a rough drawing. It launched through Google Labs in early 2026 and runs on Genie 3, DeepMind's general-purpose world model, alongside Gemini and Nano Banana Pro for prompt and image handling. The underlying organization, Google DeepMind, was formed in 2010 as an independent AI lab before merging with Google Brain in 2023. Unlike procedural generation engines that assemble pre-built assets under fixed rules, Genie 3 generates textures, geometry, physics, and unseen terrain dynamically as a user moves through the world. The system supports three core actions: sketching a new world from a written idea or picture, exploring it with walking, flying, or driving movement across first-person or isometric views, and remixing it mid-session with promptable world events such as weather changes or new objects. The model also connects to DeepMind's SIMA agent, letting researchers test goal-based navigation inside generated environments. Project Genie is best suited to game designers prototyping level concepts, AI researchers studying embodied agents, and curious Ultra subscribers experimenting with generative world-building. It is not yet a tool for shipping finished playable games: sessions have a short interaction horizon and the generated world's consistency degrades the longer a session runs. Access is gated behind Google's priciest AI Ultra subscription plan and restricted to eligible adults located in the United States; the cheaper Ultra tier leaves it out entirely. The prototype runs inside a browser with no downloads, so any desktop, tablet, or mobile device with a modern browser can use it. Google has positioned Project Genie as an early signal that frontier AI is moving past text and static video generation into real-time interactive world generation, a layer DeepMind expects to feed future gaming, simulation, and agent-training tools.
Pricing
Bundled exclusively inside Google's $200/month AI Ultra plan. The lower-cost $100/month Ultra tier, plus anyone outside the US or under 18, gets no access, and there is no separate free trial.
Key Features
- World Sketching: Converts a written idea, a photo, or a hand-drawn sketch into an explorable 3D scene rendered at 720p resolution.
- Real-Time Rendering: Generates the environment live at 24 frames per second as the user moves, rather than loading a pre-rendered scene.
- Promptable World Events: Lets users change weather, add objects, or otherwise edit the generated scene mid-session without restarting the world.
- Multi-Perspective Navigation: Supports first-person and isometric views with walking, flying, and driving movement through the generated world.
- SIMA Agent Integration: Feeds generated environments to DeepMind's SIMA agent for goal-based navigation research, linking world generation to embodied-agent testing.
- 11B-Parameter World Model: Runs on Genie 3, an 11-billion-parameter autoregressive transformer that predicts next-frame physics, textures, and terrain live rather than loading pre-built assets.
Pros
- Runs entirely in a browser tab with no installs, unlike world-model research demos that require local GPU setup.
- Accepts three separate input modes (text, image, sketch), versus most rival video-to-world tools that take only text or image.
- Ships a direct SIMA agent hook, giving researchers a path from generated world to embodied-agent testing that consumer video tools lack.
Cons
- Each session's interaction horizon runs about 60 seconds before spatial consistency starts to fray, per early Ultra tester reports.
- Access is limited to Google's priciest AI Ultra subscription tier and to eligible adult users in the United States, shutting out cheaper plans and most international testers.
- The action set is narrow: users can move and look around but cannot yet manipulate objects freely, and testers report noticeable input latency.
Data Handling
- Training-data policy
- Trained on internet video and Google's own content pipelines; DeepMind has not published a customer-data opt-out policy specific to Project Genie
Frequently Asked Questions
How much does Project Genie cost in 2026?
Project Genie carries no separate price tag of its own. It comes bundled with Google's priciest AI Ultra subscription, currently $200 a month, while the cheaper $100 a month Ultra plan leaves it out entirely. There's no pay-per-use or standalone purchase option.
Is Project Genie free to use?
No. Project Genie has no free tier or trial of any kind. It only becomes available once you're subscribed to Google's top AI Ultra plan, and even then access is restricted to adults located in the United States.
What are the best alternatives to Project Genie?
Runway suits teams that need a finished, editable video output with motion brush and camera controls rather than a live explorable world. Kling AI is a strong pick for cinematic short clips without any interactive navigation. Pika works well for quick, shareable social video when a walkable 3D environment is not the goal.
How does Project Genie compare to Runway in 2026?
Runway renders a finished video clip you edit after the fact, with a broader creative toolkit for production pipelines. Project Genie instead generates a live, navigable 3D world that a user steers in real time, but that world only holds together for about a minute per session.
How do you get started with Project Genie?
Sign up for Google's top AI Ultra plan, confirm eligibility as a US-based adult, then open labs.google/projectgenie and sign in with the same Google account. From there, type a prompt, upload an image, or sketch a scene to generate a world to explore.
Top Alternatives
- Runway: Pick Runway if you need a production video-editing pipeline with motion brush and camera controls; pick Project Genie if you want a live, walkable 3D world instead of a rendered clip.
- Kling AI: Pick Kling AI if you're a creator who wants finished cinematic video output; pick Project Genie if you want to steer and explore a world in real time as an embodied user.
- Pika: Pick Pika if you need fast, shareable short-form video clips; pick Project Genie if you need a persistent, interactive environment you can navigate for close to a minute.