Segment anything from text
Describe any object in natural language and get a precise mask.
SAM 3 RESEARCH
SAM 3
Segment anything with text and visual prompts.
Use language and visual examples to precisely identify, segment, and follow any object in images or videos.
person
backpack
INTERACTIVE DEMO
Prompt a concept. See every instance.
Try a text prompt or add a visual example to explore SAM 3.
SAM 3Text promptVisual promptImage + text
Add a visual exampleDrop an image to guide the concept
Upload a videoTrack objects across frames
Mode Segment + trackOutput Precise masks
SAM 3 IN ACTION
From words to visual concepts.
TEXT: “PERSON” · 98%
Open-vocabulary segmentation
Find concepts beyond a fixed label set.
Multi-instance detection
Separate every matching object at once.
Video tracking
Follow concepts as scenes change.
Visual understanding
Explore fine-grained real-world objects.
KEY CAPABILITIES
Understand the visual world.
Use visual exemplars
Point to an example image when language alone is not enough.
Detect every instance
Find all matching objects in a crowded scene.
Follow objects in video
Track concepts through motion, occlusion, and camera changes.
Understand fine-grained concepts
Handle long-tail, detailed, and real-world categories.
Build visual datasets
Turn open-vocabulary prompts into scalable annotations.
HOW IT WORKS
Prompt. Segment. Track.
Prompt
Describe a concept or show an example.
Segment
Get a precise mask for every instance.
Track
Follow the concept through your video.
CHOOSE YOUR WORKFLOW
One model, many ways to explore.
WorkflowPromptResult
Image segmentationText or visual examplePixel-precise masks
Video trackingConcept + videoConsistent masks over time
Dataset creationOpen-vocabulary promptScalable annotations
USE CASES
Built for the next generation of visual tools.
Creative editing
Robotics and embodied AI
Visual search
Dataset creation
Research workflows
Video understanding
SAM 3 WORKSPACE
Explore with credits.
Start with a flexible pack and use every SAM 3 workflow.
Explorer
500credits$9 one time
Creator
2,000credits$19 one time
Research
5,000credits$39 one time
Team
12,000credits$79 one time
FAQ
Questions about SAM 3.
What is SAM 3?+
SAM 3 is a unified model that uses text and visual prompts to identify, segment, and follow objects in images and videos.
What can I use as a prompt?+
Use natural language, a visual exemplar, or both together to define the concept you want to find.
Can SAM 3 segment multiple objects?+
Yes. SAM 3 can detect and segment every matching instance in an image or video.
Does SAM 3 work with video?+
Yes. It follows prompted concepts across frames while maintaining consistent masks.
Who is SAM 3 for?+
Creators, researchers, robotics teams, dataset builders, and anyone working with visual data.
START EXPLORING