Game designers
Explore environment concepts, traversal ideas and interactive beats before building a playable implementation.
Genie 4 turns a scene idea into an environment you can explore and reshape. Begin with text or an image, then guide movement and events as the world unfolds with synchronized sound.
Explore official world recordings, then use an editable starting prompt to create your own scene.
Genie 4
Crane in the snow
Genie 4
A stained-glass landscape
Genie 4
First-person alpine skiing
Genie 4
Rowboat among water lilies
Genie 4
Backstage at the theatre
Genie 4
A river of pink silk
Google DeepMind’s world model for responsive audiovisual environments.
Google DeepMind introduced Genie 4 on November 18, 2026, followed by the developer API release on November 19. It extends the Genie world-model family with native audio and more flexible real-time steering.
A world model predicts how a scene changes in response to actions. Genie 4 combines Gemini Omni Flash with Gemini reasoning to help maintain continuity while the viewer moves and directs events.
Build a scene around these supported inputs and controls.
These are generated visual environments, rather than exported 3D meshes or deterministic game projects. Object permanence, precise physics and long-session consistency remain research challenges.
Explore a scene through motion, sound and responsive visual changes.
Move through generated spaces with keyboard and mouse controls. The view develops in response to your actions.
Use live text instructions to alter weather, introduce a character or guide an event while a world is running.
Native audio adds ambience and dialogue, making sounds part of the world rather than a separate editing step.
Combine text with an image to establish appearance before exploration begins.
Official demonstrations range from documentary scenes to stained-glass environments, with each prompt setting the visual direction.
Try cause-and-effect ideas in generated environments. Responses are probabilistic and should be evaluated rather than treated as scripted guarantees.
Use explorable scenes to develop ideas before committing to detailed production.
Explore environment concepts, traversal ideas and interactive beats before building a playable implementation.
Investigate camera paths, atmosphere and scene changes during visual development and previsualization.
Turn a style direction into a space to explore, from natural landscapes to handcrafted-looking worlds.
Investigate varied visual environments and action responses while accounting for drift and imperfect physical accuracy.
Start with a focused scene and use the generator settings to create a video variation.
Choose an example or write a setting, subject and one clear action. Include sound when it matters to the scene.
Review the selected model, available inputs, resolution and credit cost in the generator before submitting.
Generate a result, inspect its motion and sound, then adjust your prompt or settings for the next variation.
Published characteristics of Google DeepMind’s Genie 4 world model.
| Developer | Google DeepMind |
| Model type | Real-time interactive world generation |
| Developer release | November 19, 2026 |
| Inputs | Text, reference image and real-time user actions |
| Video | 720p, up to 24 fps, 16:9 |
| Audio | Synchronized 48 kHz stereo |
| Live controls | WASD / arrow navigation and mid-session text events |
| Scene continuity | Short sessions; details and object permanence may drift |
| 3D file export | Not listed as a supported output |
Last updated: November 20, 2026
Compare published world-model capabilities across two generations.
| Capability | Genie 3 | Genie 4 |
|---|---|---|
| Introduction | August 5, 2025 | November 18, 2026 |
| Interactive video | 720p at 24 fps | 720p at up to 24 fps |
| Native audio | Not listed in the launch announcement | Synchronized audio and dialogue |
| Scene steering | Navigation and promptable world events | Navigation and live language steering |
| Continuity | A few minutes in published demonstrations | Short sessions; drift remains a limitation |
Genie 3 Advantages: An established reference for interactive world generation and consistency over a few minutes.
Genie 4 Advantages: Adds native audiovisual interaction and more flexible language-guided scene changes.
Choose Genie 3 demonstrations as a visual world-model baseline; explore Genie 4 when sound and live scene direction are central to the idea.
Choose a Picir AI plan for your creative workflow. The generator displays the credit cost for the selected model and settings before you submit.
Understand its inputs, interaction style and practical limits.
Genie 4 is Google DeepMind’s interactive world-generation model. It produces audiovisual environments that respond to movement and language instructions.
Explore the Genie 4 AI Video Generator & World Model page on Picir AI. Browse the recordings and use a starting prompt to develop your own scene idea.
A world model continually predicts a scene in response to actions. A conventional video generator creates a clip from a request; an exported clip does not provide a navigable live simulation.
Google lists text, images and real-time user actions. Text or an image establishes the scene, while navigation and live instructions guide its development.
Yes. Its published output includes synchronized audio, with environmental sound and character dialogue among the demonstrated capabilities.
The published output is video with synchronized audio. A generated world should not be treated as a downloadable 3D mesh, scene graph or deterministic game project.
Visual details and object permanence can drift, especially over longer sessions or repeated complex edits. Accurate physics, legible text, spatial fidelity and audio accuracy still need improvement.
The button fills the generator with an editable starting prompt and compatible settings. It does not submit a generation. Review the selected model and cost before creating a result.