Marble, a multimodal world model that generates 3D worlds from text, images, and videos, is now publicly available
On November 12, 2025 local time,
Marble
https://marble.worldlabs.ai/
Marble: A Multimodal World Model | World Labs
https://www.worldlabs.ai/blog/marble-world-model
Spatial intelligence is the new frontier of AI, and powerful world models are essential to realizing its full potential. World models must reconstruct, generate, and simulate 3D worlds and allow both humans and agents to interact with them. 'Spatially intelligent world models will transform a wide range of industries in the coming years,' World Labs states.
World Labs released a preview of Marble, a world modeler that allows users to create 3D worlds from images and text, in September 2025. Since then, Marble has been available to early beta users, who have been creating their own 3D worlds.
World Labs announced that Marble will be made available to the public on November 12, 2025 local time.
At its public release, Marble is massively multimodal, allowing users to generate 3D worlds from text, images, videos, and 3D layouts. Users can interactively edit, extend, and combine 3D worlds, and export them as glass splats, meshes, and videos, giving users fine-grained control over how they create and edit their 3D worlds.
World Labs explains about the world model, 'Humans are inherently multimodal, integrating vision, hearing, touch, language, and more to build a mental model of the world. These different representations interact, enrich, and reinforce each other, enabling us to understand the world and act within it. The world model should work in the same way, incorporating all available input signals in a massively multimodal way into a complete 3D world, and iteratively updating its understanding of the world as new information becomes available. Marble is a unique next-generation world model that makes great strides toward realizing this vision. It now allows users to build 3D worlds from a variety of input types, and allows users to iteratively edit and extend the 3D world. Marble's new features give you full control over the generated world, allowing you to quickly create complete 3D worlds from simple images or text prompts, and interactively edit both 2D and 3D to bring the image in your head to life.'
Additionally, World Labs has launched Marble Labs, a creative hub where artists, engineers, and designers can push the boundaries of world modeling and showcase bold ideas, real-world workflows, and new possibilities in a variety of fields, including games, VFX, design, and robotics. Marble Labs also features in-depth case studies, tutorials, and documentation, enabling anyone to learn, build, and share their own 3D worlds.
Marble Labs can be accessed at:
Marble Labs | World Labs
Additionally, World Labs also announced Chisel, an AI-native tool for directly editing 3D worlds in Marble. Chisel is a new experimental editing mode for advanced users creating 3D worlds. It allows users to lay out the rough structure of their 3D world in 3D using rough 3D shapes like boxes and planes, or by importing existing 3D assets into the scene.
Chisel allows you to lay out a rough 3D scene, then add text prompts that describe the visual style of the scene, as well as any additional elements that aren't present in the rough layout. Marble then combines these inputs to create a fully detailed 3D world.
In addition, the 3D worlds that can be generated in Marble have been expanded, allowing you to generate larger 3D worlds than ever before. Also, a Composer mode has been introduced that allows you to combine 3D worlds together, allowing you to freely create larger 3D worlds than ever before.
One of Marble's export formats, Glass Splat, represents a 3D scene as a large collection of semi-transparent particles that can also be rendered in the browser using Spark , an open-source cross-platform renderer integrated into THREE.js.
You can get a good idea of what kind of 3D world you can create with Marble by watching the video below.
Welcome to Marble! - YouTube
Related Posts:
in AI, Posted by logu_ii