Category: Image tools
Remove Background From Image
Local U2Netp segmentation, previewed and downloaded as a transparent PNG
Cut out the background of one image locally with a self-hosted U2Netp model, then download a full-resolution transparent PNG. Your photo is never uploaded.
The first run downloads a small (about 5 MB) background removal model, cached by your browser for next time.
The photo and the AI model both stay on this device: nothing is uploaded, and no image content is ever logged or stored.
Everything on this page is processed in your browser. Nothing is uploaded.
What this tool does
This tool removes the background from an image inside your browser. U2Netp foreground segmentation runs through ONNX Runtime Web and WebAssembly; on first use it downloads a self-hosted model of about 4.6 MB from the same origin, and everything after that is local. Inference produces a 320 by 320 mask that is scaled back over your original-resolution image, so the preview shows both the source and the transparent result and the download is a full-resolution PNG with an alpha channel. Input is one PNG, JPEG or WebP file up to 30 MB and 24 million decoded pixels; unsupported formats and oversized images are rejected. There is no manual mask brush, edge refinement, background replacement or batch processing, no server-side processing or image upload, and no output format other than PNG. Perfect hair, fur, glass and other complex edges are not promised. First use needs a network connection unless the page runtime and model were already cached. Your image is never uploaded to LocalTools.
How to use it
- Choose one PNG, JPEG or WebP image up to 30 MB and 24 million decoded pixels.
- On first use a ~4.6 MB model is fetched from the same origin, then U2Netp segments the foreground locally.
- Compare the source and transparent previews, then download the full-resolution PNG.
Privacy
This tool runs entirely in your browser. Your input is never uploaded, stored or shared — closing the tab removes it.
Frequently asked questions
- Does it upload my image to a server?
- No. Decoding, inference and the PNG export all happen in your browser. The only network request is for the same-origin model file, never for your photo.
- How do I make a background transparent?
- Select one image, let the local model detect the foreground, then download the result. Transparency lives in the PNG alpha channel, so keep the file as PNG rather than re-saving it as JPEG.
- Why does first use take longer?
- The first run downloads a self-hosted model of about 4.6 MB and starts the WebAssembly runtime. Later runs reuse the cached model, so offline use only works once both have been cached.
- What files and limits are supported?
- One PNG, JPEG or WebP image at a time, up to 30 MB and 24 million decoded pixels. Anything larger or in another format is rejected, and the output is always PNG.
- How accurate is it, and can I refine edges or replace the background?
- U2Netp handles clear subjects well but is not promised to be perfect on hair, fur, glass or busy edges. There is no manual brush, no edge refinement, no background replacement and no batch mode.