sam3d is Meta's research-grade single-image 3D reconstruction model that combines open-vocabulary segmentation with geometry, texture, and layout predictions to transform ordinary photos into usable 3D assets. It eliminates the need for complex multi-view or LiDAR setups by leveraging advanced AI to infer full 3D structures from a single RGB image, enabling rapid and accessible 3D modeling workflows.
Key benefits include:
- Single-image 3D reconstruction: Infers full 3D shape, texture, and layout from one RGB photo, replacing multi-view and LiDAR setups.
- Open-vocabulary segmentation: Uses SAM 3 prompts (text, points, boxes) to isolate objects, allowing targeted 3D asset creation from natural language or visual cues.
- Open ecosystem for research: Provides open-source checkpoints, inference code, and curated datasets like Artist Objects for reproducible research and production pilots.
- XR-ready integration: Directly feeds AR/VR pipelines, enabling instant rendering in virtual rooms, mixed reality scenes, and immersive storytelling.
- Efficient input handling: Works with legacy photos, user-generated content, and single product shots, reducing capture complexity and hardware requirements.
Perfect for creators, AR developers, and researchers needing efficient 3D asset generation from single images while optimizing for accessibility and integration with open-source tools.
