Marble Worlds: How World Labs Advances Spatial Intelligence

Having started with language models, AI then evolved into image and video generators. Taking another leap, World labs is using AI to generate Worlds.

World Labs has unveiled its vision for spatial intelligence, a technology which could reshape how industries approach design, simulation and virtual environments through its pioneering 'world models'.

The startup has secured US$1bn in funding from major technology investors including AMD, Autodesk, Emerson Collective, Fidelity Management and Research Company, NVIDIA and Sea, signalling strong industry confidence in its approach to artificial intelligence (AI).

"This investment reflects a belief I hold deeply: the next chapter of AI will be defined by systems that can understand and reason about the physical world, not just language," says Andrew Anagnost, President and CEO at Autodesk.

"Designing and making the world requires AI that understands space, structure, physics, materials and time.

"World Labs is taking a distinct approach to AI that can create and reason about realistic 3D environments that persist and evolve.

"Combined with Autodesk's domain expertise, we're forging a different path for next generation design, simulation and digital twin workflows."

The company's first commercial offering, Marble, centres on enabling users to generate and manipulate 3D environments through multiple input methods, potentially transforming workflows across sectors from entertainment to robotics.

Creating navigable environments from simple inputs

World Labs launched Marble as a multimodal AI tool designed to generate 3D worlds from various inputs including text, images, video or basic 3D layouts.

The platform allows users to create, expand and combine spatial environments through what the company describes as AI-native world editing capabilities.

These editing tools can perform modifications ranging from minor adjustments such as repositioning objects to major transformations including visual style changes and structural reorganisation of entire environments.

The platform also offers single-step expansion functionality, enabling users to scale specific areas, objects or details within generated worlds.

"World Labs believes that spatial intelligence is the linchpin technology for creativity, for experiences, for simulation and embodied AI," says Fei-Fei Li, CEO and Co-Founder of World Labs.

The technology could support applications across cinematic production, gaming, visual effects, design and robotics sectors.

According to the company's blog, early adopters from creative and technical backgrounds are already exploring use cases spanning interactive experiences, robotics simulations and therapeutic environments.

Enabling developer integration through API

To facilitate broader adoption among development communities, World Labs has released a World application programming interface (API) which allows spatial creation capabilities to be integrated into programmable workflows through code-based implementation.

The API accepts multimodal inputs including text, single images, multi-image sets, 360° panoramas and videos to generate navigable environments that capture spatial elements such as layout, depth, lighting and structure.

Generated worlds can be exported to third-party industry tools, potentially enabling interoperability across existing technology stacks, or deployed as foundations for interactive applications.

"Marble is just a step on our journey toward spatial intelligence," the company says. "Going forward, a key opportunity is interactivity. Future world models will let humans and agents alike interact with generated worlds in new ways, unlocking even more use cases in simulation, robotics and beyond."

The company's emphasis on interactivity suggests potential evolution toward systems where both human users and autonomous agents could engage with generated spatial environments, though the practical implementation timeline for such capabilities remains unspecified.