Meta has released SAM 3, the latest version of their Segment Anything Model, which revolutionizes video segmentation by allowing users to segment objects in videos using simple text prompts. Traditionally, rotoscoping—a highly manual and labor-intensive process—required teams of people to segment elements frame by frame. SAM 3 automates this process, completing it in seconds, making video editing and analysis much more efficient. The model is open source with open weights, meaning anyone can download and run it on their own computer or use Meta’s free online playground.

Using SAM 3 is straightforward: users can type in the name of an object they want to segment, such as “dog” or “bicycle,” and the model will highlight all instances of that object throughout the entire video, frame by frame. Alternatively, users can simply click on an object in the video, and SAM 3 will recognize and track it automatically. The model can distinguish between similar objects, like bicycles and motorcycles, and provides a list of segmented objects with labels and colors for easy management. This makes it incredibly useful for detailed video analysis and editing.

The Meta-hosted playground offers a user-friendly interface where users can upload videos or select sample videos to experiment with SAM 3’s capabilities. After uploading, users describe the objects they want to segment, and the model processes the entire video to highlight those objects. Users can add effects such as contours to the segmented objects and download or share the results directly from the platform. This makes SAM 3 accessible to a wide range of users, from hobbyists to professionals.

One of the standout features of SAM 3 is its support for templates—predefined tasks that automate common video editing needs. For example, a popular use case is pixelating license plates or faces for privacy. Users can apply a template that automatically segments license plates and applies a pixelation effect across the video in seconds. This functionality is especially valuable for video editors, animators, and content creators who need to perform repetitive tasks quickly and accurately.

Beyond video editing, SAM 3 has broad applications in fields like security, wildlife monitoring, and robotics. It can track vehicles such as trucks or cars in street camera footage, identify birds for nature observation, or help robots recognize and respond to objects and people in their environment. The model’s open-source nature and ease of use open up endless possibilities for innovation. Meta encourages users to download SAM 3, experiment with it, and explore the creative and practical applications it enables.