From image to transparent PNG.

1. Choose a supported image

Select a JPG, PNG, or WebP file up to 20 MB. The browser first checks the file type and size, then confirms that it can decode the image. Damaged or unsupported files are rejected before inference starts.

2. Load the browser model

A dedicated web worker loads a pinned background-removal model. Capable browsers try the WebGPU path; when WebGPU cannot initialize, the application can use a WebAssembly fallback. Model files are checked against expected file sizes and SHA-256 hashes before use and may be cached by the browser for later sessions.

3. Estimate the subject mask

The model works from a resized inference input and returns an alpha mask. The application maps that mask back to the original image dimensions, applies the pipeline’s built-in edge refinement, and combines it with the original pixels.

4. Review and download

The result opens in a local editor where you can compare the original and removed views, preview several background colors, zoom, and download a transparent PNG. Preview background colors are not baked into the downloaded file.

What “local” means

Your image pixels are not uploaded to bgremoved.pro or to a third-party image-processing API. Normal requests still leave the browser to load the website, model files, fonts, and other application assets. See the privacy policy for the exact distinction.

Known limitations

Automatic segmentation can struggle with contact shadows, reflections, transparent objects, fine strands, low contrast, and subjects that visually merge with the background. Always inspect important outputs at full size.

For a repeatable review process, use the cleaner-results checklist.