Meta has announced the open release of SAM 3 (Segment Anything Model 3), the most advanced evolution of its segmentation model family.
This model redefines how objects are identified and tracked in images and video: now a simple text prompt or example image is enough for the system to locate all instances that match the description.
Unlike previous versions, which required clicks or bounding boxes to delimit objects, SAM 3 understands natural language, enabling instructions such as “people with backpacks,” “red cars,” or “bicycles on the right.”
The model also maintains stable temporal consistency, allowing it to follow moving objects even when the shot or lighting changes.
What makes SAM 3 different
🔹 Text- or image-guided segmentation
You can describe what you want to find or upload a reference image. SAM 3 automatically identifies all matches.
🔹 Video tracking
It generates consistent masks across frames—essential in dynamic environments.
🔹 Detection + tracking in a single system
One model performs both tasks, simplifying computer-vision pipelines.
🔹 Fully open source
Meta released the code and weights, enabling research, deployment, and enterprise integrations.
Real-world use cases and applications of SAM 3
- Counting people in queues or events
Identifies and counts individuals without custom models. - Tracking moving vehicles
Highly useful for mobility, access control, traffic analytics, or smart parking. - Automatic face blurring
Ensures privacy in corporate videos, public recordings, or datasets. - Automatic license-plate blurring
Regulatory compliance with no manual work. - Smart editing and post-production
Manipulate objects with prompts like “blue chair” without manual cutouts. - Retail and e-commerce
Product detection, shelf analytics, and interactive experiences. - Science, sports, and technical analysis
Precise tracking of animals, particles, athletes, or machinery.
Availability: SAM 3 is now on Azure Microsoft Foundry
Microsoft has added facebook-sam3 to the official Microsoft Foundry model catalog, accessible from the Hugging Face registry.
This means companies can use SAM 3 directly on Azure, with no need to deploy it manually in containers or on their own GPUs.
Official resources
- Repository (code + weights):
👉 https://github.com/facebookresearch/sam3 - Official playground:
👉 https://aidemos.meta.com/segment-anything/gallery
Bravent is already integrating SAM 3 in Azure
The arrival of SAM 3 on Azure opens the door to advanced solutions for video analytics, automation, anonymization, and computer vision, fully integrated with cloud infrastructure.
At Bravent we are already testing its potential in real use cases such as people counting, vehicle tracking, and intelligent privacy.
María Soto Castro
Head of Innovation - BraventIf you want to explore how SAM 3 can be applied in your organization, we can help you with a PoC or full integration.
Contact us at info@bravent.net and discover how to bring this innovation to your organization.
