AI Background Remover

How to remove a background

1. 1 Pick a quality mode. Fast uses a small 5 MB model and answers in seconds; Precise downloads a 170 MB model once and handles hair, fur and fine edges much better.
2. 2 Add a photo (JPG, PNG or WebP). The model runs on your own device; nothing is uploaded.
3. 3 Wait for the cutout; the first use of a mode includes its one-time model download.
4. 4 Check the preview and download the transparent PNG.

Frequently asked questions

Which mode should I use?

Start with Fast. If edges look rough, especially around hair or fuzzy outlines, switch to Precise; it uses a far larger model that was trained for exactly those cases.

Why is the Precise model 170 MB?

That is the size of the neural network's weights. It downloads once, your browser keeps it, and every later use starts instantly. Nothing about your images is sent anywhere.

What do I get as output?

A PNG with a real alpha channel at up to the original resolution. You can place it on any background in any editor, or feed it straight into a design tool.

Why did a solid color appear instead of transparency?

Some apps flatten PNGs on import. The file itself is transparent; check it in an editor that shows a checkerboard behind transparent areas.

Which models does this use?

Fast is U2Net-p and Precise is DIS (isnet-general-use), both open models under the Apache 2.0 license, running via onnxruntime in your browser.