SAM 3D

Verified

Meta's open models that rebuild 3D objects and human bodies from a single image

Visit SAM 3D →No rating yetFree

Product preview

SAM 3D

What it is

SAM 3D is a research release from Meta that extends the Segment Anything family into 3D. It has two models. SAM 3D Objects takes a single image, lets you select an object, and produces a posed, textured 3D reconstruction you can view from other angles, typically within seconds according to Meta. SAM 3D Body estimates 3D human pose and body shape from one image, handling unusual poses, partial occlusion and multiple people, with optional guidance from masks or keypoints. Anyone can try both in Meta's Segment Anything Playground in the browser, and model checkpoints and inference code are published on GitHub.

Key features

Single-image 3D object reconstruction
3D human pose and body estimation
Browser playground for testing
Open checkpoints and code on GitHub

What we liked

+ Free to try in the browser
+ Open models for developers and researchers
+ Works from just one photo

Worth noting

· Moderate output resolution
· Reconstructs one object or person at a time

Editor's notes

More context from AI BirdView

Getting a 3D model from an ordinary photo usually requires multiple angles or a dedicated scanning app. SAM 3D shows how far single-image reconstruction has come, and Meta has made it easy to explore by offering a browser playground alongside open code. Its pairing of object and human body models makes it useful for a wide range of experiments.

SAM 3D Objects turns a selected object in one photo into a textured, posed 3D model.
SAM 3D Body estimates human pose and body shape, even with occlusion or several people in frame.
Masks or 2D keypoints can guide the body model for more control.
The Segment Anything Playground lets anyone upload images and test results without installing anything.
Checkpoints and inference code are available on GitHub for research and development work.

SAM 3D is best for developers, researchers, AR and VR builders and curious 3D hobbyists who want to experiment with single-image reconstruction. A practical tip: start in the browser playground with a clear photo where the object is fully visible, and review the license in each GitHub repository before using the models in a commercial project.