SAM Image Matting
Build a subject mask with points and brush controls when automatic background removal needs more guidance. The interactive workflow is useful for isolating a specific person or object in a busy scene.
Core capabilities
- Positive and negative point prompts
- Brush and eraser refinement
- Undo and redo history
- Background color export
How it works
The Segment Anything workflow first encodes the uploaded image into reusable visual features. Positive and negative clicks then tell the model what should be included or excluded, producing a mask that can be refined before export.
Interactive prompts make this tool better suited to ambiguous scenes than a one-click remover. The user remains responsible for checking the mask, especially around narrow gaps and fine edges.
How to use SAM Image Matting
- Upload a supported image or choose an example
- Left-click areas to include and right-click areas to exclude
- Use brush, eraser, undo and redo to refine the mask
- Choose a background color or export the cutout
Practical uses
- Selecting one object from a scene containing several subjects
- Preparing cutouts for background replacement
- Refining masks with points, brush strokes and erasing
Limitations
SAM predicts object regions rather than creating a professional hand-traced path. Very fine hair, translucent material and tiny disconnected details may need manual cleanup. Model loading can use substantial memory on older devices.
Privacy and processing
Image encoding and prompt-based mask generation are designed to run in the browser. The initial model download is relatively large, so first use depends on network speed even though the subsequent inference is local-first.
Frequently asked questions
What is SAM image matting?
It uses the Segment Anything model with user prompts to estimate the object region that should be kept.
What do left and right clicks do?
A left click marks content to include; a right click marks content to exclude.
Can I refine the generated mask?
Yes. Brush, eraser, undo and redo controls are available for additional cleanup.
Are images processed locally?
The core feature encoding and mask workflow are designed to run in the browser after model resources load.
AI Creator tools run in modern desktop browsers.