# Meta SAM 3

> Open-source vision model for zero-shot object segmentation and tracking in video

Meta SAM 3 (Segment Anything Model 3) is a computer vision model designed for object detection, segmentation, and tracking across video frames. It allows users to isolate objects using natural language prompts without manual keyframing or rotoscoping. The model is released for mask-generation tasks and supports video-wide tracking and effect application.

Website: https://huggingface.co/facebook/sam3

## What it is for

- Track and isolate objects across full video sequences using text prompts
- Automate video rotoscoping for post-production and VFX pipelines
- Apply automated effects like blurring or pixelation to specific moving subjects

## Features

- Zero-shot object detection and segmentation from natural language prompts
- Consistent object tracking across sequential video frames
- Support for automated mask generation via Hugging Face Transformers
- Integration with video effects including blur, pixelate, glow, and outlines

## Pricing

Open source. Free and open-source (custom license)

## Details

- Type: AI model
- Category: [AI](https://xiry.app/topics/ai.md)
- Open source: yes
- Platforms: Code library, Web, Self-hosted
- For: Machine learning researchers, computer vision developers, and video editors
- Alternative to: Segment Anything Model (SAM), SAM 2
- Tags: computer vision, segmentation, video tracking, object detection, meta, huggingface

## Why it was saved

- [@rpn, 19 November 2025](https://www.instagram.com/reel/DRPsIeJCXlg/): Eliminates manual rotoscoping by allowing users and developers to isolate, track, and apply effects (blur, pixelate, mask, outline) to any object across entire video clips automatically.

## Related

- [SAM 2](https://xiry.app/resources/sam-2): Promptable visual segmentation model for images and real-time video tracking. (Web app, Open source)
- [askcomputer](https://xiry.app/resources/askcomputer): AI camera assistant interface built around the iOS Dynamic Island (Web app)
- [Chatterbox](https://xiry.app/resources/chatterbox): Open-source zero-shot text-to-speech and voice cloning model (AI model, Open source)
- [datasocial/tiktok-5.6B-videos](https://xiry.app/resources/datasocial-tiktok-5-6b-videos): Public dataset of 5.6 billion TikTok video metadata records in Parquet format (Freemium)
- [EgoX](https://xiry.app/resources/egox): AI framework generating first-person egocentric videos from third-person video footage (AI model, Open source)
- [LivePortrait](https://xiry.app/resources/liveportrait-623): Efficient portrait animation from still photos guided by driving videos. (AI model, Open source)

Updated 3 October 2026. Page: https://xiry.app/resources/meta-sam-3

Make your own library: Xiry takes your saved reels, tweets, Reddit bookmarks and Watch Later videos, makes them searchable and pulls the useful stuff out, like this library. It's invite-only for now; join the waitlist at https://xiry.app/#waitlist.