Vi3W
Skip to content

Guides

SAM 3D Objects vs Body vs Align: Meta's SAM 3D models explained

Meta's SAM 3D Objects, SAM 3D Body and SAM 3D Align explained: what each does, its limits, and how to use them free in your browser with Vi3W.

3 min read

A yellow armchair lifting out of a framed photo of a cluttered living room and becoming a 3D object

Quick answer

What is SAM 3D?

Meta's family of models reconstructs 3D from a single ordinary photo. SAM 3D Objects lifts a chosen object out of a busy scene, SAM 3D Body estimates a person's 3D pose and body shape, and SAM 3D Align places the people and objects back into one shared scene. All three run in Vi3W.

Why SAM 3D is different

Most image-to-3D models want a clean picture: one object, centred, on a plain background. Real photos are not like that. Meta built SAM 3D for the messy case. It was released on November 19, 2025, and Meta's announcement describes it as bringing common-sense 3D understanding to everyday images.

Vi3W now runs the three SAM 3D models: Objects, Body and Align. SAM 3D Objects is on every Vi3W plan, including free.

SAM 3D Objects: pull one thing out of a photo

You start from an ordinary photo, pick the object you want, and SAM 3D Objects reconstructs its shape, texture and position as a posed 3D model. It is the right first try when:

  • the object is in a room, on a shelf or on a street;
  • other things partly hide it;
  • you don't want to cut the background out by hand.

Meta lists three limits. Output resolution is moderate, so complex objects lose fine detail. Whole people can come out distorted. And it reconstructs objects one at a time without reasoning about contact between them. For a clean studio shot that needs maximum detail, a model like TRELLIS 2 or Hunyuan 3D 3.1 Pro will usually do better.

SAM 3D Body: pose and body shape of a person

SAM 3D Body estimates a person's 3D pose and body shape from a single image, including unusual poses and people partly hidden by other things. It can be guided with segmentation masks and 2D keypoints. Its output is based on Meta's Momentum Human Rig, which separates the skeleton from the body's soft-tissue shape.

Meta's stated limits: it processes each person separately, without modelling interactions between people or with objects, and its hand poses are not as accurate as specialised hand-tracking models. It works on one image, so it is not a replacement for motion capture.

SAM 3D Align: put it all back together

SAM 3D Align takes the original photo, a body mesh from SAM 3D Body and, optionally, an object mesh from SAM 3D Objects. It works out scale, position and rotation from the photo, so the person and the objects end up in the same 3D scene without manual lining up. The fal.ai model page lists its outputs: an aligned body mesh in PLY and GLB, and a combined scene GLB when you include an object.

A typical use: someone sitting in a chair. Run Body on the person, Objects on the chair, then Align to seat them together.

Which SAM 3D model to pick

  1. An object in a real-world photo: SAM 3D Objects.
  2. A person's pose and body shape: SAM 3D Body.
  3. A person and objects together in one scene: Body, then Objects, then Align.
  4. A clean product shot that needs fine detail: try another model; see the full list.

If your result looks wrong, our guide to why image-to-3D looks wrong covers the usual causes.

Try SAM 3D in Vi3W

Log in to app.vi3w.in and upload any photo. SAM 3D Objects is included on the free plan, and every result can be exported as GLB, FBX or OBJ.

Frequently asked questions

sam 3d vs trellis 2

SAM 3D Objects is built for real photos where the object sits in a scene. TRELLIS 2 does best with a clean image of one object, and gives full PBR materials and thin open surfaces. Busy photo: SAM 3D. Clean cut-out: TRELLIS 2.

sam 3d to blender

Generate in Vi3W, export GLB or FBX, then use File > Import in Blender. SAM 3D Align also outputs GLB, so a combined scene of a person and objects imports in one go.

sam 3d body motion capture

SAM 3D Body estimates pose and shape from a single image, not motion over time, so it is not a motion-capture system. It gives you a posed body mesh for one frame.

Last reviewed