DocumentationSAM 3 Video Segmentation
SAM 3 Video Segmentation is a powerful tool that allows you to easily track and segment objects across video frames. Whether for creating visual effects, removing backgrounds, or analyzing video content, this tool provides pixel-perfect accuracy.
Try SAM 3 Video Segmentation Now →
Features
- Automatic Tracking: Define your object in the first frame, and AI precisely tracks it through all subsequent frames.
- No Green Screen Needed: AI identifies and isolates subjects regardless of background complexity.
- Multiple Interaction Modes: Support for text descriptions (Quick Mode) or AI Agents (Natural Language Understanding) to lock onto targets.
- Cloud GPU Acceleration: Complex video processing is completed in cloud clusters, ensuring fast response times.
How to Use
1. Upload Video
Click the upload area or drag and drop files to get started.
- Supported Formats: MP4, WebM, MOV.
- Limitations: Currently supports videos up to 50MB and 30 seconds in length to ensure optimal processing efficiency.
2. Define the Target
You can use two modes to tell the AI what you want to track:
Quick Mode
- Enter English keywords in the input box (e.g.,
dancing person,red cup). - This mode is ideal for clear targets with direct descriptions.
AI Agent Mode
- Switch to the AI Agent tab.
- Select a keyframe and describe your target in natural language (e.g.,
Identify the black car that moves from left to right). - Click Analyze this Frame to preview the identification results.
3. Processing and Preview
- Analysis Preview: Before starting full-video processing, you can see how the AI identifies objects in your selected frame.
- Execute Processing: Once the target is confirmed, click Process Video. The AI will extract embeddings frame-by-frame and generate segmentation paths.
- Track Progress: You can view the processing status in real-time, including format optimization, merging, and audio restoration.
4. Download Results
After processing is complete, you can:
- Preview Results: Play the processed preview video directly in your browser.
- Download Video: Save video files with segmentation masks or isolated subjects.
Credits
- Video processing is billed based on duration, typically consuming 1 credit per second of video.
- The system will display an estimated cost in real-time based on your video's length.
FAQ
Do I need to manually label every frame?
No. You only need to define the target in the first frame or via text description. SAM 3's detector-tracker architecture handles the rest automatically.
Do prompts have to be in English?
Yes, to ensure the highest accuracy for object recognition, we recommend using English descriptions for both AI Agent and Quick Mode.
What if my video is too long?
If your video exceeds 30 seconds, we recommend trimming it first. The 30-second limit is currently in place to ensure a stable and fast processing experience for all users.