Atlas
AI model · Robotics / world model · Text → Video by World Labs
Turn images into explorable 3D worlds with spatial intelligence.
Atlas is a multimodal world model for creators and researchers that transforms images, text, and video into explorable 3D worlds. It enables professional-grade scene reconstruction, grayboxing, and mesh conversion for use in film, architecture, and robotics.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Launched
- 2026-11-12
- Languages
- English
Category: Generative AI
- Released
- 2026-11-12
- Input Modalities
- Text, Image, Video, Panorama, Multi-view
- Output Formats
- 3D Worlds, Meshes, Video Renders, Gaussian Splats
- Integration Support
- Blender, NVIDIA Isaac Sim, MuJoCo, RoboSuite
- Core Tools
- Viewfinder, Chisel, Splats to Mesh, Relight
- Access Method
- World API, Web Platform
- Turns images into explorable 3D worlds
- Converts Gaussian splats into meshes with real geometry
- Supports text, image, panorama, and video inputs
- Integrates with professional tools like Blender
- Enables real-time camera path rendering and scene editing
Overview
Atlas is World Labs' next-generation world model designed to perceive, generate, and reconstruct 3D environments from multimodal inputs. It serves businesses and creators in industries such as Film & VFX, Architecture, and Robotics by turning static images or text prompts into explorable, editable worlds. The model features specialized tools like Viewfinder for scene reconstruction and Chisel for controllable grayboxing and structural design. Atlas distinguishes itself by enabling the conversion of Gaussian splats into meshes with real geometry, making generated worlds compatible with professional 3D tools and game engines. It is accessible through a developer API and integrates directly with industry-standard software like Blender.
Perception and Generation
- Scene Reconstruction: The Viewfinder tool reconstructs 3D scenes from photos or generated images, allowing users to explore new viewpoints.
- Camera Path Rendering: Users can design and render specific camera paths through the reconstructed 3D environments.
- Controllable Scene Building: The Chisel tool enables users to block out a scene to guide its structure and spatial layout through grayboxing.
- Dynamic Scene Shaping: Atlas allows users to move objects within a scene and see every camera view update in real time.
Simulation and Interaction
- Splat-to-Mesh Conversion: The model converts Gaussian splats into meshes with real geometry for use in interactive environments.
- Lighting Manipulation: Atlas provides tools to explore new ways to light and manipulate splats within the generated worlds.
- Simulation Readiness: The model generates physically plausible environments suitable for integration into simulators like NVIDIA Isaac Sim and MuJoCo.
Workflow and Integration
- Blender Integration: Atlas capabilities can be integrated into existing creative tools, including Blender.
- Developer API Access: The World API allows developers to generate explorable 3D worlds from text, images, panoramas, and video.
- Cross-Platform Export: Generated worlds can be rendered directly in web browsers or exported to downstream production tools.
Use cases
- Filmmakers build virtual sets from concept art and render cinematic camera paths for production.
- Architects transform sketches into explorable 3D environments to walk through designs with clients.
- Robotics researchers generate thousands of diverse, physically plausible simulation environments for training autonomous agents.
- Game developers convert AI-generated scenes into meshes with real geometry for interactive gameplay.
- Immersive media creators turn 2D cinematic sequences into navigable 3D worlds for social storytelling.
Specifications
- Modalities
- Text, images, panoramas, multi-view inputs, video → 3D worlds, meshes, videos
- Domains
- Film & VFX, Architecture & Design, Robotics, Gaming, Immersive Media
- Access
- Developer API, web-based tools (Viewfinder, Chisel), and integrations for Blender
- Released
- 2026-11-12
- Input Modalities
- Text, Image, Video, Panorama, Multi-view
- Output Formats
- 3D Worlds, Meshes, Video Renders, Gaussian Splats
- Integration Support
- Blender, NVIDIA Isaac Sim, MuJoCo, RoboSuite
- Core Tools
- Viewfinder, Chisel, Splats to Mesh, Relight
- Access Method
- World API, Web Platform
How it works
- Input and Reconstruction: Users upload photos, generated images, or text prompts to the Viewfinder tool to reconstruct a scene.
- Scene Shaping and Grayboxing: The Chisel tool allows for blocking out structures and moving objects to guide the spatial layout of the world.
- Refinement and Conversion: Users can manipulate lighting and convert splats into meshes to prepare the environment for production.
- Workflow Integration: The final 3D world is exported to tools like Blender, Unreal Engine, or robotics simulators for final use.
FAQ
What types of input can I use with Atlas?
Atlas supports a wide range of inputs including text prompts, single images, multi-image sets, 360° panoramas, and video.
Can I export Atlas worlds to other 3D software?
Yes, Atlas includes a Splats to Mesh tool that converts Gaussian splats into meshes with real geometry for use in games and 3D tools.
Which industries is Atlas built for?
Atlas is designed for professional workflows in Film & VFX, Architecture & Design, and Robotics research.
Is there an API available for Atlas?
The World API allows developers to generate worlds on demand and embed them directly into their own products and workflows.
Added by AntSpace from public sources (worldlabs.ai).