Image Segmentation

Segment images into labeled regions using semantic segmentation models. Useful for background removal, scene understanding, and image editing.

For full API reference (segmentImage(), options, result types, and custom providers), see the Core Vision guide.

See it in action

Run it in the Background Remover block — an installable, registry-composed version that runs segmentation entirely in the browser.

Recommended Models

Model	Size	Speed	Use Case
`briaai/RMBG-1.4`	~176MB	⚡⚡	High-quality background removal (non-commercial license)
`Xenova/segformer-b0-finetuned-ade-512-512`	~90MB	⚡⚡	Semantic segmentation (MIT license)

Background Removal Example

Based on the Background Remover block:

import { transformers } from '@localmode/transformers';
import { segmentImage } from '@localmode/core';

const model = transformers.segmenter('briaai/RMBG-1.4');

async function removeBackground(imageDataUrl: string) {
  const { masks } = await segmentImage({
    model,
    image: imageDataUrl,
    abortSignal: controller.signal,
  });

  // Find the main subject mask (e.g., 'person', 'wall', etc.)
  // Apply the mask to create a transparent background
  return applyMaskToImage(imageDataUrl, masks);
}

Best Practices

Segmentation Tips

License — RMBG-1.4 is non-commercial only; commercial use requires a paid license from BRIA AI. Use segformer-b0 (MIT) for commercial projects.
Purpose-built for background removal — RMBG-1.4 excels at foreground/background separation
Multiple masks — Each segment in the image gets its own mask and label
Combine masks — Merge related masks for background removal (e.g., combine all non-subject masks)
Canvas processing — Use Canvas API to apply masks for transparency effects

Composed Block

Block	Description	Links
Background Remover	SegFormer segmentation to a transparent-background PNG	Live · Install: `npx shadcn add @localmode/ui/blocks/image-studio/background-remover`

Next Steps

Core Vision API

Full API reference including segmentImage().

Object Detection

Detect objects with bounding boxes.

Image Features

Extract feature vectors for similarity search.