Sam 3D Body

SAM 3D Body (3DB): Single-Image 3D Human Mesh Reconstruction

Edited by Segmind Team on December 12, 2025.

What is SAM 3D Body (3DB)?

SAM 3D Body (3DB) is an advanced 3D reconstruction model that generates complete human body meshes from just a single image. Developed by Facebook Research, it is built on the Momentum Human Rig (MHR) parametric framework, empowering it to generate precisely detailed full-body 3D meshes that include hands, feet, and body poses, even when parts of the person are hidden or captured from complicated angles.

SAM 3D Body (3DB) goes beyond basic pose estimation models by delivering meshes ready for production, making them perfect for animation workflows, AR/VR experiences, and realistic digital human creations. Furthermore, its encoder-decoder design takes regular RGB photos as input and offers the option to implement 2D keypoints or segmentation masks as additional guidance for improved accuracy and quality.

Key Features of SAM 3D Body (3DB)

• •Full-body mesh recovery: It is capable of generating complete 3D meshes, including fingers, toes, and facial structure, from a single image.
• •Occlusion handling: It can accurately reconstruct body parts even when they are partially hidden or obstructed.
• •Multi-person support: It effectively processes group photos and outputs individual or combined meshes.
• •Flexible input options: The model works with standard images and accepts optional 2D keypoint or mask prompts.
• •Robust viewpoint handling: It proficiently maintains accuracy across diverse camera angles and perspectives.
• •Real-world trained: It is built on high-quality, diverse datasets reflecting uncontrolled environments

Best Use Cases

• •

  Gaming and Animation: It is perfect for generating rigged character meshes from reference photos for rapid prototyping or background character creation.

• •

  AR/VR Development: It can create realistic avatars from selfies for immersive experiences and virtual try-on applications.

• •

  Motion Capture Alternative: It can be used to reconstruct 3D poses from event photography or archival images where traditional mocap wasn't available.

• •

  E-commerce & Fashion Tech: It can be used to build virtual fitting rooms by generating customers' body models from uploaded photos.

• •

  Film and Visual Effects: It can quickly create digital doubles or crowd simulations from reference photography.

Prompt Tips and Output Quality

• •

  Image Selection: Well-lit photos with clearly visible subjects render the best output. Additionally, full-body shots generate perfectly complete meshes, and though SAM 3D Body (3DB) handles cropped images carefully for high-quality results.

• •

  Multiple People: Enable the return_individual_meshes option to process group photos as it will yield separate mesh files for each person that the model detects in the images; it is specifically important for applications that nedd individual character manipulation.

• •

  Keypoint Integration: Toggle include_keypoints when you need 2D/3D joint positions for rigging, analysis, or pose verification to add skeletal data to the output without affecting mesh quality.

• •

  Occlusion Scenarios: The model performs exceptionally well in scenarios with partial occlusions (i.e., people are behind furniture, overlapping poses); though in the case of a high level of occlusions, the output may be approximated results for hidden parts.

SAM 3D Body (3DB) Effective Usage Guide

SAM 3D Body (3DB) excels at turning a single RGB image into a full-body 3D mesh. Below are best practices, sample parameter settings and prompt tips to get optimal results in different scenarios.

1. Image Selection

• Use well-lit, high-resolution photos showing as much of the body as possible.
• Avoid extreme motion blur, heavy shadows or lens flares that obscure joints.
• For studio-style shots, a plain or low-contrast background helps the model delineate edges.

2. Core Parameters

• include_keypoints (bool, default true):
– Set to true when you need 2D/3D joint coordinates for rigging, animation or pose analysis.
– Turn off for faster inference if mesh alone is sufficient.
• return_individual_meshes (bool, default true):
– Enable for group photos when you need separate mesh files per person.
– Disable to combine all detected subjects into one mesh (useful for crowd-style outputs).

3. Use-Case Settings

1. •

   Rapid Prototyping (Single Character)
   • include_keypoints: false
   • return_individual_meshes: false
   • Notes: Faster turnaround when you only need a single combined mesh.

2. •

   Character Rigging & Animation
   • include_keypoints: true
   • return_individual_meshes: false
   • Notes: Joint data simplifies skinning and motion retargeting.

3. •

   Group Photos or Crowd Simulation
   • include_keypoints: false
   • return_individual_meshes: true
   • Notes: Separate outputs per actor let you assign unique materials or animations.

4. •

   Virtual Try-On & Avatar Creation
   • include_keypoints: true
   • return_individual_meshes: true
   • Notes: Full rig plus individual meshes provide maximum flexibility for AR/VR.

4. Prompt Tips

• •Crop margins tightly around subjects to focus detection.
• •When body parts are occluded (crossed arms, furniture), the model will infer missing geometry—but results improve with partial visibility (e.g., one arm at least).
• •Group shots: ensure subjects are not too close together to avoid mesh merging.

5. Performance & Quality

• •Inference speed is similar with or without keypoints; toggling mainly affects output size.
• •Default settings balance accuracy and speed—tweak only if your pipeline demands specialized data.
• •Always review mesh normals and topology in your 3D tool; minor cleanup may be required for complex poses.

Use this guide to configure SAM 3D Body (3DB) for your project needs and unlock fast, production-ready 3D human reconstructions from any single image.

Is SAM 3D Body (3DB) open-source?

SAM 3D Body (3DB) by Facebook Research has been released with publicly available code, pretrained checkpoints, and demo notebooks for experimentation and application development.

What file formats does 3DB output?

The SAM 3D Body (3DB) model generates 3D mesh data compatible with standard graphics pipelines. It supports individual mesh exports while processing multiple subjects.

How does 3DB handle hands and feet compared to other models?

Compared to most of the body mesh models that approximate extremities, 3DB uses MHR representation to capture detailed hand and foot articulation, making it superior for applications that need finger-level precision.

Can I process images without full-body visibility?

Yes, the model reconstructs available body parts and intelligently creates occluded regions; the best results can be achieved from images showing most of the body.

What parameters should I adjust for best results?

Start with default parameters: enable include_keypoints if you need skeletal data for rigging; use return_individual_meshes for processing groups, and you need separate character files.

How does 3DB compare to Mediapipe or OpenPose?

Mediapipe or OpenPose excel at 2D keypoint detection; 3DB generates full volumetric mesh geometry suitable for 3D rendering, animation, and spatial computing applications.