It’s now possible to create immersive 3D worlds from a single photograph.
That’s according to World Labs, Fei-Fei Li’s new startup, which just unveiled its advanced AI system that generates immersive three-dimensional scenes directly within a web browser.
The San Fransisco-based startup, valued at over $1 billion at less than a year old, says the tool democratizes content creation. The AI extracts depth, texture, and spatial data from flat images, transforming them into dynamic 3D environments. The tool extracts depth, texture, and spatial data from flat images, using advanced AI to interpret visual cues and recreate 3D worlds.
The AI learns by processing massive datasets of 2D images and 3D models, mastering how visual features create realistic spatial relationships. The imagery used for World Labs’ demonstration were generated using FLUX 1.1, Ideogram, and Midjourney. They also used public domain artwork by Vincent van Gogh, Edward Hooper, Georges Pierre Seurat, and Wassily Kandinsky.
Potential use cases are wide-ranging, including creating realistic environments quickly for gaming, immersive learning for schools, and virtual tourism. In manufacturing, the tool could aid in factory layout planning, process optimization, and worker training. The tech could also benefit robotics, providing realistic 3D simulations for training autonomous systems, optimizing robot navigation in complex settings, and testing designs in lifelike visual scenarios before real deployment.
The spatial intelligence startup emerged from stealth mode earlier in 2024 with $230 million in funding. They’re developing so-called Large World Models that move beyond current AI models that only handle 2D images and text. They want to make AI tools that allow users like artists, designers, and engineers to create and edit 3D worlds with realistic physics and details.
To do this, AI needs a lot of data from various sources like 3D scans of environments, videos, and images from the real world. It would also use data about how objects behave and interact from simulations and virtual environments.