AI · AI model
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.
What it's 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
- Topic
- Computer Vision
- 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: 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
Promptable visual segmentation model for images and real-time video tracking.
AI camera assistant interface built around the iOS Dynamic Island
Open-source zero-shot text-to-speech and voice cloning model
Public dataset of 5.6 billion TikTok video metadata records in Parquet format
AI framework generating first-person egocentric videos from third-person video footage
Efficient portrait animation from still photos guided by driving videos.
Updated