Meta's open-source AI model that segments and tracks any object in images and videos with a click, box, or mask prompt, in real time.
About
Meta Segment Anything Model 2 (SAM 2) is an AI segmentation model from Meta that selects and tracks any object in an image or video with fast, precise results. It addresses a long-standing challenge in computer vision: segmenting objects consistently across video frames, including when objects are occluded or leave the frame, without requiring task-specific training data.
SAM 2 is the first unified model for object segmentation across both images and videos. Users select an object with a simple prompt—a click, a bounding box, or a mask—and the model then tracks that object throughout the video using a streaming memory architecture that processes frames one at a time. Meta released SAM 2 as an open-source project, along with the SA-V video segmentation dataset, a demo, and code, so researchers and developers can build on the work directly.
Key Features
Pricing & Fees
- Pricing: SAM 2 is offered as an open release; no paid plans are described. Not publicly listed on the pages we reviewed; check the official site for current plans.
- Free availability: The pretrained model, code, demo, and SA-V dataset are downloadable at no cost from Meta's official site and GitHub.
- Training dataset: The SA-V dataset contains roughly 600K+ masklets across about 51K videos, collected in real-world scenarios spanning 47 countries, with annotations for whole objects, parts, and occlusions.
- Performance: Meta reports SAM 2 outperforms existing video object segmentation models—notably for tracking object parts—and improves on SAM for image segmentation while requiring less interaction time than prior interactive methods.
- Extensibility: Segmentation outputs can feed other AI systems such as video generation models, and the input prompt types can be extended in future work.
Who It's For
SAM 2 is aimed primarily at computer vision researchers, machine learning engineers, and developers building tools that need precise object selection or tracking in images and video. Practical use cases include video editing and effects, dataset annotation, augmented and live-video interactions, and pipelines where segmentation output feeds downstream models like video generators. Because the model and dataset are open source, academic teams and independent developers can experiment with it without licensing barriers, while product teams can integrate it as a component in larger AI systems.
FAQ
What is Meta Segment Anything Model 2?SAM 2 is Meta's open-source AI model that segments any object in images and videos from a click, box, or mask prompt and tracks it across video frames in real time.
Is SAM 2 free to use?Yes, Meta publicly releases the pretrained model, the SA-V dataset, a demo, and code at no cost; no commercial pricing tiers are described on the official site.
How does SAM 2 track objects that disappear from view?A per-session memory module records information about the selected object from earlier frames, so the model can resume tracking it when it reappears.
How does SAM 2 differ from the original SAM?SAM 2 extends SAM's promptable segmentation to video with a streaming memory architecture while improving image segmentation results; on images, the memory is empty and it behaves like SAM.
What is the SA-V dataset?SA-V is a large, geographically diverse video segmentation dataset of about 51K videos and 600K+ masklets from 47 countries, which Meta open sourced alongside the model.
Related Tutorials
View all →
Related Tools
View all →
Qwen AI - Multimodal Assistant Qwen AI is a versatile assistant that handles text, images, and more, providing comprehensive support. 
Hugging Face Hugging Face is an open community platform where machine learning practitioners host, discover, and collaborate on over two million models, datasets, and AI applications. 
Llama Llama 4 is a powerful AI model family with Scout and Maverick versions. Experience advanced AI capabilities. 
LTX 2 AI Video Generator - Create Stunning Videos Free Online Transform your ideas into amazing AI-generated videos with LTX 2 Video Generator. Create professional videos online for free. No experience required. Start generating now. 
Mistral AI - Open AI Alternative Mistral AI provides enterprises with customizable AI platforms for deploying assistants, agents, and multimodal AI using open models.