Meta SAM 3: The New Era of Segment Anything
Next-Gen Segment Anything<br>Meta SAM 3 - Segment Anything with Text, Clicks & Concepts
Meta SAM 3 lets you point, type, or show an example and it instantly finds, segments, and tracks every object you care about in any image or video.
Get Started Now
Learn More
Download
Overview<br>What Is Meta SAM 3 AI?
Meta SAM 3 AI (Segment Anything Model 3) is Meta AI Research’s newest “segment anything” model that can detect, precisely segment (mask), and track objects in images and videos using simple prompts.
What makes SAM 3 special is how you can tell it what to segment:
Key Values and Vision
Text prompts: type a short phrase like “person” or “car” and it masks matching objects.
Exemplar prompts: draw a box around one example object, and it masks all similar objects in the scene.
Visual click prompts: use positive/negative clicks to include/exclude parts of an object.
Interactive refinement: if it misses something, add follow-up prompts to improve the mask.
Video tracking: once an object is segmented, it can follow that object through video frames.
Meta has also said SAM 3 is designed for real-world tools, including upcoming uses in Instagram Edits and Meta AI app features like Vibes, plus research collaborations (wildlife and underwater imagery).
Why Choose Meta Sam 3
Because it makes "segment anything" fast and flexible letting you mask and track objects in images and videos using simple text prompts, exemplar boxes, or click-based guidance, with easy refinements when results need fine tuning.
Get Started Now
Multiple prompt types
Segment using text, exemplar box, or positive/negative clicks.
“Segment anything” flexibility
Works across many object types without needing fixed labels.
Video support
Tracks segmented objects through videos (not just single images).
Easy refinement
Add follow-up prompts to quickly fix missed areas or mistakes.
Fast creative workflows
Great for effects, cutouts, background removal, and object highlighting.
Built on SAM 2 strengths
Keeps proven features while adding stronger text + exemplar segmentation.
Evolution<br>Evolution of Meta SAM From SAM 1 to SAM 3 (and SAM 3D)
SAM 1 - The "one-click" foundation (Images)
Segment any object in an image with as little as a single click
Improve results with follow-up clicks (refinement)
Best for: fast object cutouts in still images
Read more
SAM 2 - Segmentation + tracking (Images + Video)
Segment and track objects in videos, not just images
Supports click, box, or mask prompts
Still includes refinement with extra clicks
Best for: video masking workflows where the object moves over time
Read more
SAM 3 - Unified “segment anything” with language + exemplars (Images + Video)
Keeps SAM 2 capabilities (click prompts, refinement, tracking)
Adds open-vocabulary text prompts - Mask objects using short phrases (e.g., object categories described in words)
Adds exemplar prompts - Box one example → segment all matching objects
Built as a unified promptable model across images and video
Best for: fastest real-world workflows when you want to guide segmentation by text, examples, or clicks
Read more
Features<br>Key Features of Meta SAM 3 AI
Segment Anything in Images & Video: Precisely masks objects in both photos and videos.
Object Tracking in Video: Tracks segmented objects across frames for consistent results.
Text Prompts (Open-Vocabulary): Use short words/phrases to segment all matching objects.
Exemplar Prompts: Draw a box around one example object, and SAM 3 segments all similar objects.
Visual Prompts (Clicks): Use positive + negative clicks to include or exclude parts of an object.
Interactive Refinement: Add follow-up prompts to correct mistakes and improve mask accuracy.
State-of-the-Art Performance: Strong results across text + visual segmentation tasks.
Unified Promptable Architecture: One model supports language + exemplars + visual prompts across images and video.
Real-World Ready Workflows: Designed for creator and research use cases (effects, editing, analysis).
SAM 3D Extension: Supports future 3D reconstruction/analysis for spatial understanding.
More Features
Quick Start<br>How to Get Started with Meta SAM 3
You can start experimenting with SAM 3 in a browser playground, via hosted APIs, or by running the official open source code on your own GPU.
Input an Image or Video
You start by uploading an image or video to SAM 3 (e.g., in the Segment Anything Playground or your own application).
Choose a Prompt Method
Text Prompts
Exemplar Prompts
Visual Click Prompts
Model Understands the Scene
Behind the scenes, SAM 3 uses a powerful vision encoder to analyze the image/video and understand object boundaries, textures, colors, and more.
Generate Segmentation Mask
Based on your prompt, the model creates a pixel accurate segmentation mask identifying the object(s) you've specified.
Interactive Refinement
If SAM 3 misses part of the object or...