SAM 3, by Meta

What is SAM 3?

SAM 3 — Segment Anything Model 3 — is an AI model built by Meta that can instantly identify and isolate any object in an image or video. You click on something (or describe it), and SAM 3 draws an exact mask around it: the outline of a person, the edge of a product, the shape of a tree against a sky.

It is the third generation of Meta's Segment Anything project, first released in 2023. SAM 3 extends the original by working on video in real time — not just individual photos — and handling difficult edges like hair, fur, glass, and semi-transparent objects far more accurately.

This is not an image generator. SAM 3 does not create images. What it does — selecting arbitrary objects with a single click — is something that previously required a skilled designer and several minutes of careful masking per image. SAM 3 does it in milliseconds.

What does "segmentation" mean?

Segmentation means dividing an image into distinct regions. A mask is the precise pixel-level outline that separates one object from the rest of the image.

Before AI segmentation tools, isolating an object required:

• The pen tool in Photoshop — tracing every edge by hand
• Green screen / chroma key — only worked in controlled studio conditions
• Background removal APIs — worked well for simple cases (person on plain background) but struggled with complex scenes

SAM 3 changes this by understanding the structure of any image and producing precise masks for any object, in any scene, with no training on that specific image type.

How to actually use SAM 3

Meta has released SAM 3 as open-source research — there is no official Meta app for general consumers. You access it in one of three ways:

Option 1 — Try it in the browser (no setup)

Several tools have integrated SAM 3 or its predecessor into their online editors:

• Segment Anything demo — Meta's own web demo. Upload an image, click on objects, download masks.
• Adobe Photoshop — The "Remove Background" and "Select Subject" tools in Photoshop use SAM-style segmentation under the hood.
• Canva — Background Remover uses similar technology for one-click subject isolation.
• Remove.bg — Dedicated background removal tool powered by segmentation AI.

Option 2 — Use it via an image editing app

Many professional tools now ship SAM 3 or compatible models as a built-in feature:

• DaVinci Resolve 19 — Magic Mask tool uses SAM for video object isolation
• GIMP (with plugin) — Open-source alternative to Photoshop
• ComfyUI / Automatic1111 — Stable Diffusion interfaces include SAM nodes for precise inpainting masks

Option 3 — Run it locally (technical)

If you want to run SAM 3 directly:

1. Install Python and PyTorch on your machine
2. Clone the SAM 3 repository from Meta's GitHub (facebookresearch/sam2)
3. Download model weights (Meta provides several sizes — the smaller ones run on most modern GPUs)
4. Use it via Python scripts or integrate it into a workflow with tools like ComfyUI

This route gives you the most control and is free with no usage limits. It requires a GPU — an NVIDIA card with 8GB+ VRAM is recommended for real-time video use.

What makes SAM 3 better than SAM 2 or SAM 1?

SAM 3 is significantly better at tracking objects through video — where earlier versions might lose track of an object when it moves, rotates, or gets partially obscured, SAM 3 maintains accurate masks across frames.

Real-world use cases

E-commerce product photography — Upload 50 product images, run SAM 3 to remove backgrounds in bulk, output clean PNGs on white. What used to take a designer hours now takes minutes, and the results are cleaner than manual masking for most products.

Video editing — Isolate a person from their background in a talking-head video. Apply a different background, blur the original, or apply color grading only to the subject. SAM 3 tracks the person frame-by-frame without a green screen.

Social media content — Cut yourself out of a photo to place on a branded background. Create cutout-style graphics where the subject overlaps text.

Architecture and real estate — Replace skies in exterior property photos, isolate interiors, swap out furniture for virtual staging.

AI image workflows — When using Stable Diffusion for inpainting (editing specific parts of an image), SAM 3 creates the precise mask that tells the AI exactly which region to modify. Much faster than drawing masks by hand.

Data labelling — If you're building a computer vision model and need to annotate thousands of images, SAM 3 auto-generates masks that humans only need to review and approve rather than draw from scratch.

SAM 3 vs other background removal tools

Use SAM 3 if you need precise control, video support, or want to integrate segmentation into a custom workflow or app. Use Remove.bg or Canva if you want a zero-setup browser tool for quick background removal on people and products.

The bigger picture

SAM 3 is a foundational model — a building block that other tools use rather than a finished product in itself. Its impact is already visible everywhere: the "Remove Background" button in your phone's photo app, the subject isolation in Instagram Stories, the one-click masking in modern video editors — all of these are powered by the same class of segmentation AI that SAM pioneered.

Meta releasing it as open source means any developer can integrate it into any product for free. The practical result: precise, instant object selection is now a standard feature rather than a premium add-on.