Today, we are pleased to announce Echo-1, a frontier model for 3D world generation. Our model takes text or an image as input and generates a 3D-consistent world.
Interactive World Exploration
Echo-1's main strength is that it can take a simple text prompt or an image and generate a rich, 3D-consistent world from it. This makes it useful for creative 3D scene generation, digital twinning pipelines, and 3D design workflows. Imagine turning a picture of a factory floor into a structured 3D representation, or describing "a flexible exhibition space with sculptural lighting" and seeing it appear as an explorable environment. Echo-1 turns both ideas and real places into interactive 3D spaces.
Echo-1 worlds are interactive in real time. Once the model generates a 3D representation, users have complete control over the camera. They can explore the new world from any angle, and each new view renders instantly. This real-time rendering works even on low-end hardware. You don't need expensive, specialized equipment, so many more people can create and explore 3D worlds, from professional designers to casual users.
Generated Scenes to Explore
Below are several generated scenes that you can explore freely. Move through each world to see the quality of the 3D generation for yourself.
How It Works
Echo-1 uses a simple pipeline. It begins with an input image or, when starting from text, a generated reference image. From this input, the model predicts a physically grounded 3D representation of the scene, capturing both geometry and appearance in a consistent spatial layout. We then convert this internal representation into a format suited to real-time exploration. For the web demo, we use 3D Gaussian Splatting (3DGS), which renders very fast on a GPU and lets you view scenes interactively in the browser, even on modest hardware.
Scene Representation Based on 3D Geometry
Echo-1 generates a scene representation based on 3D geometry. The model infers one consistent 3D structure for the whole scene that encodes both geometry and appearance at metric scale. This means every rendered view, and its depth map, comes from one consistent 3D model of the scene. The current version uses 3D Gaussian Splatting (3DGS) to render the scene because it is very efficient, but the representation itself is flexible, and we can convert it into other formats as rendering technology changes.
We believe this accurate 3D structure is essential for many real-world applications such as digital twinning or content creation. For example, Echo-1 can turn a single photograph of a construction site or manufacturing cell into a 3D reconstruction at real-world scale, so teams can plan maintenance, simulate layout changes, or check space limits without rescanning the site.
For content creation, consistent 3D worlds let teams iterate faster in graphics applications such as gaming (e.g., generating large game levels), and they can also provide more realistic training data for robotics (e.g., geometrically accurate simulation environments for navigation and manipulation). Because Echo-1 keeps the whole scene consistent by design, these workflows stay reliable even with very sparse input from everyday devices, or when the only input is a text prompt.
Spatial Understanding and Editing
Echo-1 introduces a set of capabilities that make 3D world generation immediately useful across creative, design, and professional workflows.
Scene Restyling
With built-in scene restyling, users can restyle generated or reconstructed environments. This is useful for interior redesign, real estate visualization, and game level restyling. Users can quickly explore variations such as "minimalist Scandinavian," "industrial loft," or "warm, rustic wood tones," without manually rebuilding the scene.
Explore different styles of the same scene in 3D.
Spatial Understanding
Echo-1 can also make more localized scene edits. To do this, it helps to first split the scene into its parts. Echo-1 generates a semantic segmentation mask of the scene that identifies parts such as "chair", "table", "floor", or "wall".
Scene Editing
Echo-1 can use a text prompt, or the scene breakdown described above, to remove, add, or replace objects in the scene. Below are examples of scenes with these edits. You can explore the edited worlds interactively:
Object Removal
Echo-1 can also remove selected objects from a cluttered scene to show a clear view of the space behind them.
Future Plans
Echo-1 is only the first step in a much larger roadmap. We are building toward a future in which anyone can generate, modify, and reason about complex 3D environments with natural language. Upcoming versions will support full scene editing from text prompts, so users can add, remove, rearrange, or restyle objects just by describing what they want.
Further ahead, our model will add dynamics and physical reasoning on top of the underlying representation. Scenes will then look consistent and also behave according to physics, which makes interactive simulations, robotics testing, and richer digital twin applications possible. Echo-1 is the starting point. The next versions will be more capable World Models that change, respond, and interact like their real-world counterparts.
You can explore the generated scenes on our website, move through the examples, and try the model's capabilities in real time. If you want to try Echo-1 on your own data or workflows, .