SAM 3D Objects [image-to-3d]
SAM 3D Objects reconstructs 3D geometry and texture from a single image at $0.02 per generation. Trading multi-view requirements for single-image convenience, it handles real-world clutter and occlusion that trip up traditional photogrammetry. Built for product visualization, AR content pipelines, and rapid 3D prototyping where you need dimensional assets from existing photos.
Built for: E-commerce 3D catalogs | AR/VR asset creation | Physical object digitization | Game asset prototyping
Single-Image 3D Reconstruction Without the Studio Setup
SAM 3D Objects uses visual context recognition, not just geometric patterns, to infer complete 3D structure from partial views. Where traditional reconstruction demands controlled lighting and multiple camera angles, this model works with everyday photos containing occlusion, shadows, and background clutter.
What this means for you:
- Context-aware segmentation: Auto-isolate objects using text prompts ("chair", "lamp"), point clicks, or bounding boxes with no manual masking required for clean extractions
- Multi-object scene handling: Process multiple objects per image with individual Gaussian splats and GLB meshes, plus combined scene files for unified workflows
- Flexible depth integration: Upload external point map data (NPY/NPZ format) to override automatic depth estimation when you have better source data from LiDAR or structured light
- Production-ready outputs: Generates both Gaussian splat (.ply) files for rendering and GLB meshes for game engines, with per-object metadata for transformation pipelines
SAM 3D Objects delivers production 3D assets at $0.02 per reconstruction, enabling 50 generations per dollar on fal.
| Metric | Result | Context |
|---|
| Cost per Generation | $0.02 | 50 reconstructions per $1.00 on fal |
| Output Formats | Gaussian splat (.ply) + GLB mesh | Dual format for rendering and game engine workflows |
| Segmentation Methods | Text/point/box prompts + manual masks | Auto-segmentation or precision control via mask URLs |
| Multi-Object Support | Individual + combined outputs | Separate files per object plus unified scene reconstruction |
Technical Specifications
| Spec | Details |
|---|
| Architecture | SAM 3D Objects |
| Input Formats | Single image (JPG, PNG, WebP, GIF, AVIF) + optional mask URLs/pointmap data |
| Output Formats | Gaussian splat (.ply), GLB mesh, metadata JSON, optional artifacts ZIP |
| Segmentation Options | Text prompt, point coordinates, bounding boxes, or manual mask URLs |
| Related Endpoints | SAM 3D Align for full scene reconstruction |
| License | Commercial use permitted |
API Documentation | Quickstart Guide
How It Stacks Up
SAM 3D Align – SAM 3D Objects focuses on precise object-level reconstruction with flexible segmentation controls for extracting specific elements from complex scenes. SAM 3D Align enables full scene reconstructions, placing objects and humans in a shared context together for immersive spatial experiences.
Tripo3D Image-to-3D – SAM 3D Objects prioritizes visual context understanding for cluttered real-world images at $0.02 per generation, using segmentation-driven reconstruction to handle occlusion and scene complexity. Tripo3D delivers production-ready stylized 3D models at $0.20-$0.40 depending on texture quality, optimized for clean-input workflows where background isolation is already handled.
Hunyuan3D v2 – SAM 3D Objects extracts and reconstructs specific objects from complex scenes using SAM-based segmentation at $0.02 per generation. Hunyuan3D v2 generates complete 3D assets from single images at $0.017 per generation, focusing on versatile asset creation with native 3D generation rather than reconstruction from real-world photos.
Hyper3D Rodin – SAM 3D Objects handles real-world photography with occlusion and clutter at $0.02 per generation with dual Gaussian splat and GLB output. Hyper3D Rodin generates production-ready 3D models with PBR materials at $0.40 per generation, supporting both image-to-3D and text-to-3D workflows for CG-friendly assets.
