3D vision research
Evaluate single-image reconstruction on cluttered, occluded, and otherwise difficult real-world scenes.
Independent tool overview
Sam 3D at a glance
SAM 3D is Meta's pair of research models for reconstructing textured objects, scene layout, and full-body human meshes from a single image.
Visit the official Sam 3D site ↗
- Developer
- Meta Superintelligence Labs
- Model pair
- SAM 3D Objects and SAM 3D Body
- Input
- A single image plus an object mask or optional body prompts
- Output
- Textured object reconstructions or articulated human meshes
- Access
- Free web demo and gated downloadable checkpoints
- License
- Meta SAM License; review its use and redistribution conditions
Overview
What Sam 3D is
SAM 3D is not one general-purpose 3D generator. It is a paired release: SAM 3D Objects reconstructs selected objects with shape, texture, pose, and scene layout, while SAM 3D Body estimates an articulated human mesh, including the body, feet, and hands. Both are designed for difficult natural images rather than only isolated product shots.
The easiest way to understand the release is Meta's Segment Anything Playground. Developers and researchers can also request model checkpoints and run the published inference code locally. The models are useful for reconstruction, research, prototyping, AR previews, robotics perception, and motion or pose analysis, but their output still needs inspection before production use.
Use cases
Who Sam 3D is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
AR and spatial prototypes
Turn a photographed object into a reconstruction for view-in-room, camera exploration, or early interaction tests.
Robotics perception experiments
Estimate object shape, pose, and layout from natural images as one component in a larger perception system.
Human pose and shape analysis
Recover a promptable full-body mesh from one photo, including hands and feet, using SAM 3D Body.
Game and media previsualization
Create reference reconstructions from photographs before an artist performs topology, cleanup, rigging, and optimization.
Capabilities
Core Sam 3D features
Single-image object reconstruction
SAM 3D Objects infers full object shape, appearance, pose, and layout from one selected region in a photograph.
Natural-scene focus
The model is designed for small objects, indirect views, occlusion, clutter, and other conditions common outside studio images.
Multi-object workflow
Published notebooks show how to process multiple masked objects and place their individual reconstructions in a common scene.
Gaussian-splat export
The official object inference example saves its result as a PLY Gaussian splat for downstream viewing or processing.
Promptable human recovery
SAM 3D Body estimates a full human mesh and can accept segmentation masks and 2D keypoints as guidance.
Momentum Human Rig
The body model uses Meta's MHR representation to separate skeletal structure from soft-tissue shape.
Combined body and object example
Meta provides a notebook for aligning SAM 3D Body and SAM 3D Objects outputs in the same frame of reference.
Open inference code
Meta publishes repositories, setup instructions, example notebooks, and checkpoint download links for both model families.
Web playground
The Segment Anything Playground offers a lower-friction way to select people or objects and inspect a reconstruction.
Research assets
The release includes papers, evaluation materials, a 3D Artist Object Set, and a public leaderboard for comparing object reconstruction work.
Process
How the Sam 3D workflow works
Step 1
Choose the correct model
Use SAM 3D Objects for physical objects and scene elements; use SAM 3D Body for an articulated human reconstruction.
Step 2
Prepare the image
Use a clear source you are allowed to process, then identify the target with a mask or supported prompt.
Step 3
Try the playground
Test a few representative images in Meta's web experience before committing time and compute to a local setup.
Step 4
Request and install checkpoints
For local work, accept the model terms, obtain gated Hugging Face access, and follow the official Python and GPU setup.
Step 5
Run and export
Execute the appropriate notebook or demo, inspect the camera and pose, and export the reconstruction in the supported format.
Step 6
Validate and refine
Check hidden surfaces, scale, collisions, hand pose, topology, texture, and scene alignment before using the asset downstream.
Cost
Sam 3D pricing and free plan
Meta does not sell SAM 3D as a metered commercial API. The playground, code, and model checkpoints are available without a listed subscription price, but local use requires compatible hardware and is governed by the SAM License.
Segment Anything Playground
Free
A hosted way to experiment with SAM 3D on uploaded images.
- No paid plan is listed
- Designed for interactive experimentation
- Review Meta's current playground terms before uploading sensitive or third-party images
Local model release
Free download; infrastructure costs apply
Run the published code and gated checkpoints in your own environment.
- Checkpoint access must be requested through Hugging Face
- Use and redistribution are subject to the SAM License
- GPU, storage, engineering, and hosting costs are your responsibility
Pricing checked . Check current pricing at the source ↗
Assessment
Sam 3D strengths and limitations
Where it stands out
- Separates object reconstruction from specialized articulated-human recovery.
- Targets clutter, occlusion, unusual viewpoints, and other real-world conditions.
- Reconstructs object pose and scene layout rather than producing only an isolated concept asset.
- SAM 3D Body supports interactive guidance with masks and 2D keypoints.
- Meta publishes inference code, checkpoints, notebooks, papers, and evaluation assets.
- The web playground gives non-specialists a quick way to test the capabilities.
- Local inference offers more control over deployment and image handling than a hosted-only product.
- The object and body outputs can be combined in a shared frame through Meta's example workflow.
What to consider
- A single image cannot reveal hidden geometry with certainty; unseen surfaces are model inferences rather than measurements.
- Meta says SAM 3D Objects has moderate output resolution, which can lose detail on complex targets.
- The object model predicts objects individually and does not jointly reason about contact, support, or interpenetration.
- A multi-object scene can therefore contain collisions, floating elements, or physically inconsistent relationships.
- SAM 3D Body processes each person separately and does not model interactions between multiple people, objects, and the environment.
- Meta says its hand-pose accuracy does not exceed specialized hand-only methods.
- Object reconstruction is not the same as a production-ready polygon asset; artists may still need retopology, UV work, rigging, cleanup, and optimization.
- Body estimates should not be treated as exact biometric measurements or clinical assessments.
- The downloadable setup is technical and assumes a compatible Python, PyTorch, CUDA, and GPU environment.
- Checkpoint access is gated and not available in comprehensively sanctioned jurisdictions.
- The SAM License is not a standard permissive open-source license; commercial and redistribution plans require a careful license review.
- Generated reconstructions can reproduce copyrighted designs, identifiable people, or sensitive surroundings from the source image.
- The public demo and local release do not provide a turnkey production API, SLA, asset-management system, or editing pipeline.
- Benchmark and vendor preference results do not guarantee quality on a specific image category or downstream workflow.
Compare
Sam 3D alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Design
Rodin
Choose Rodin for a commercial text-and-image-to-3D asset workflow with hosted generation and production-oriented export options.
Explore Rodin →Design
Seed 3D
Consider Seed 3D for research focused on simulation-ready scene and asset reconstruction.
Explore Seed 3D →Design
Hunyuan 3D Smart Topology (PolyGen)
Consider Hunyuan 3D PolyGen when professional polygon topology and artist-oriented asset generation are the priority.
Explore Hunyuan 3D Smart Topology (PolyGen) →Design
Meshcapade
Choose Meshcapade when the central requirement is controllable digital humans and body models rather than general object reconstruction.
Explore Meshcapade →Questions
Sam 3D FAQs
What is Meta SAM 3D?
SAM 3D is a pair of Meta research models: SAM 3D Objects reconstructs selected objects and scene layout from one image, while SAM 3D Body recovers an articulated human mesh.
Is SAM 3D the same as SAM 3?
No. SAM 3 focuses on promptable concept segmentation in images and video. SAM 3D uses a selected object or person to produce a three-dimensional reconstruction.
Is SAM 3D free?
Meta lists no subscription price for the playground, code, or checkpoints. Local inference still incurs hardware and engineering costs, and use is governed by the SAM License.
Can SAM 3D create a 3D model from one photo?
Yes. It can infer a textured object reconstruction or human mesh from one image. Geometry that is hidden in the photograph is estimated, so it should not be treated as an exact scan.
Can SAM 3D reconstruct people?
Yes. SAM 3D Body estimates the body, feet, and hands using the Momentum Human Rig and can be guided with masks or 2D keypoints.
Can I run SAM 3D locally?
Yes. Meta publishes code and gated Hugging Face checkpoints for both model families. The official setup requires a technical Python and GPU environment.
Can SAM 3D reconstruct a whole scene?
SAM 3D Objects can reconstruct multiple selected objects and their layout, but it predicts objects individually and does not fully reason about physical contact or collisions.
Is SAM 3D output ready for a game or film?
Usually not without review and additional work. Depending on the use case, an artist may need to fix geometry, topology, scale, materials, rigging, intersections, and performance.
Bottom line
Our Sam 3D verdict
SAM 3D is one of the most useful research releases for testing grounded 3D reconstruction from ordinary photos, especially because Meta provides both an interactive demo and local code. It is best treated as a reconstruction and prototyping foundation—not an automatic replacement for scanning, production asset work, or validated human measurement.
Visit Sam 3D website ↗Similar Miscellaneous Tools
View all Miscellaneous AI tools →
Mistral Studio
Mistral AI Studio - A platform for companies to move from AI prototypes to production with built-in tools for performance tracking, testing, and security.
Rnj-1
Rnj-1 - Essential AI's open-source 8B parameter model that matches or outperforms similar-sized rivals
HunyuanWorld-Mirror
HunyuanWorld-Mirror - Generate 3D worlds from text, image, or video inputs
GLM-4.6V
GLM-4.6V - Zhipu AI's open-source multimodal model family with native tool use capabilities
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.