Meta's SAM 3D Body Reconstructs Human Pose From Images

Meta AI has released SAM 3D Body, a model that reconstructs a 3D human mesh and estimates body pose from ordinary 2D images. Published on Hugging Face, the release extends Meta's Segment Anything family — long associated with 2D segmentation — into the harder territory of three-dimensional human reconstruction.

Human mesh recovery turns a flat photo into a posed, volumetric model of a person, capturing limb positions and body shape. That capability underpins a range of applications, from motion capture and avatars to AR, fitness, and robotics perception. The model name references DINOv3, Meta's self-supervised vision backbone, suggesting the system leans on strong learned image features to anchor its 3D predictions.

Why it matters

Single-image 3D reconstruction has historically required specialized rigs or multi-camera setups. Bringing it to a single photo, with an openly available checkpoint, lowers the barrier for researchers and developers who want to experiment without heavy capture infrastructure.

• Recovers full 3D body mesh from a single image
• Handles pose estimation alongside shape
• Builds on Meta's DINOv3 vision features

The model is distributed under a custom license, so teams planning commercial use should review the terms on the model card before building on it. As an initial release, it sets a baseline that Meta and the wider community can iterate on.