HCODX/Background Remover
Local-only · AI runs in your browser · No upload

Background Remover

Remove the background from any photo — entirely in your browser using the MODNet AI model via Transformers.js. Get a transparent PNG, fill with a solid color, or apply your own background. Both library and model are Apache 2.0 licensed.

Need a refine brush, custom backgrounds, or higher-accuracy model? Pro Try the AI Background Remover Pro — multi-model, brush + undo/redo, edge feather, de-spill, smart crop, image / gradient backgrounds, drop shadow.

Drop an image here

Or click to choose. JPG, PNG, WebP. Works best on portraits and product photos.

Choose image
Device: WebAssembly (CPU). First run downloads ~25 MB AI model.
Loading…
Before & after
Original
Result
Background
Fill mode
Use cases

What you'll use this for

Profile photos

Get a clean transparent or solid-color headshot for LinkedIn, Slack, or a CV.

Product photos

Cut out products for Etsy, eBay, or Shopify on a uniform white background.

Memes & collages

Drop a cutout subject onto any new background in your editor.

ID-style photos

Replace a busy room background with a regulation-friendly solid color.

Step by step

How to remove an image background

1

Drop in an image

Drag and drop or click to choose. JPG, PNG, and WebP are supported.

2

Click Remove background

The first time, a ~25 MB AI model is downloaded and cached. After that, processing takes 1–10 seconds depending on image size.

3

Pick a background

Transparent for PNG export, or pick a solid color. The preview updates live.

4

Download

PNG keeps transparency; JPG is smaller and works everywhere but requires a solid background color.

FAQ

Frequently asked questions

No. The AI model is downloaded to your browser on first use (~25 MB, cached afterward) and runs entirely on your device. The photo bytes never leave your computer.

MODNet (Xenova/modnet on Hugging Face), Apache 2.0 licensed, run through Transformers.js. It is trained primarily on portraits and people, but works well on many products and isolated objects too.

Portraits, product photos on plain or mildly-busy backgrounds, and well-lit objects work best. Very complex scenes, fine hair on patterned backgrounds, and translucent objects (glass, smoke) are the hardest cases — those usually need manual touch-up afterward.

The ~25 MB AI model is downloaded once on first use. After that it is cached in your browser and subsequent runs are fast (1–10 seconds depending on image size and device).

Yes — on browsers with WebGPU enabled (Chrome 113+, Edge 113+), processing is 3–10× faster. The tool automatically falls back to WebAssembly otherwise. You can see which backend is in use in the small label above the result preview.

Fine hair on a patterned background is the hardest case for any segmentation model — even the best paid services struggle. MODNet is a relatively small model (~25 MB) chosen for browser performance. For best results, photograph subjects on a plain background and use even lighting.

About

About browser-based background removal

This tool runs entirely on your device — no upload, no server-side processing, no privacy compromise. It uses two open-source pieces:

  • Transformers.js by Hugging Face — the JavaScript port of the popular Transformers library. Apache 2.0 licensed, runs ONNX models in the browser via WebGPU or WebAssembly.
  • MODNet — a real-time portrait matting model originally published by Sun et al. (2022). The ONNX port on Hugging Face is Apache 2.0 licensed and ~25 MB.

How segmentation works

The model takes your image, downsamples to 512×512, and predicts an alpha matte — a single-channel image where each pixel is a value from 0 (background) to 1 (foreground). We then composite the original image with this matte to produce a clean cutout. Soft edges around hair are represented as fractional alpha values, which preserves transparency gradients.

Limitations

  • Best on portraits and product photos; struggles on complex multi-subject scenes.
  • Output is 512px on the long edge inside the model, then upscaled to your image size — high-resolution photos look slightly softer at the cutout boundary.
  • Hair, fur, glass, and smoke are inherently hard. Touch up in any image editor with an alpha-aware brush.

Privacy

Because everything runs in your browser, the photo never leaves your computer. The model and Transformers.js library load from a CDN; your image bytes do not.

Related

Related tools