SAM 3 Align | [image-to-3d]
Meta's SAM 3D Alignment model unifies 3D reconstructions at $0.02 per scene, placing human bodies and objects into spatially coherent environments from single images. Trading standalone mesh generation for full scene composition, it solves the critical problem of isolated 3D assets that don't relate to each other spatially. Built for developers creating AR/VR experiences, game assets, or digital twins where context matters as much as individual object quality.
Use Cases: AR/VR Scene Composition | Game Environment Assembly | Digital Twin Creation
At $0.02 per aligned scene versus $0.05+ for sequential body-then-object workflows, SAM 3D Alignment delivers 2.5x cost efficiency by processing spatial relationships in a single inference pass rather than requiring manual composition.
| Metric | Result | Context |
|---|
| Scene Alignment | Single-pass spatial coherence | Processes body + object meshes together vs sequential workflows |
| Cost per Scene | $0.02 | 50 aligned scenes per $1.00 on fal |
| Input Flexibility | 2-4 mesh inputs | Supports body mesh (required) + optional object mesh + mask guidance |
| Output Formats | PLY, GLB, visualization overlay | Animation-ready rigged output with scene preview |
| Related Endpoints | Sam 3 Image to 3D (Body), Sam 3 Image to 3D (Objects) | Body reconstruction vs object reconstruction vs alignment variants |
Full Scene Reconstruction Without Manual Composition
SAM 3D Alignment operates as a post-processing stage after mesh generation, using MoGe depth estimation to establish shared spatial context. Unlike traditional workflows where you generate body meshes and object meshes separately then manually position them in Blender or Unity, this model calculates scale, translation, and rotation automatically from the original reference image.
What this means for you:
- Automatic spatial relationships: Upload a body mesh and object mesh from the same photo, get back a unified scene where proportions and positions match the original image, no manual adjustment of transforms or eyeballing scale ratios
- Flexible input pipeline: Accepts SAM 3D Body meshes (PLY or GLB) as required input, with optional SAM 3D Object meshes for combined scenes, plus optional mask guidance for precise body region isolation
- Multi-format output: Returns aligned body mesh in both PLY (for processing) and GLB (for immediate 3D preview), plus a visualization overlay showing mesh alignment on the original image for quality verification
- Metadata-driven accuracy: Leverages focal length from upstream SAM 3D Body metadata when available, or estimates from MoGe depth when not provided, eliminating the guesswork in camera parameter matching
Technical Specifications
| Spec | Details |
|---|
| Architecture | SAM 3D Alignment |
| Input Formats | Image URL (JPEG, PNG, WebP), Body Mesh URL (PLY/GLB), Optional Object Mesh URL (GLB), Optional Body Mask URL |
| Output Formats | Aligned Body Mesh (PLY, GLB), Scene GLB (when object provided), Visualization Overlay (PNG), Alignment Metadata (JSON) |
| Depth Estimation | MoGe-based spatial context with optional focal length override |
| License | Commercial use permitted |
API Documentation | Quickstart Guide | Enterprise Pricing
How It Stacks Up
SAM 3 Image to 3D (Body) ($0.05) – SAM 3D Alignment ($0.02) handles the spatial composition step after body reconstruction, trading mesh generation for scene-level coherence at 2.5x lower cost. Sam 3 Image to 3D (Body) remains essential as the upstream provider of body meshes that this alignment model requires as input.
SAM 3 Image to 3D (Objects) ($0.05) – SAM 3D Alignment processes pre-generated object meshes into unified scenes rather than creating them from scratch. Sam 3 Image to 3D (Objects) generates those object meshes initially, making it the complementary first step before alignment when building multi-element scenes.
SAM 3 Image to Image ($0.01) – SAM 3D Alignment ($0.02) extends SAM 3's segmentation foundation into full 3D spatial reasoning for scene composition. SAM 3 Image to Image provides the 2D segmentation and masking that can guide body region isolation during alignment workflows.