Skip to main content
Free Image Tool

Background Remover

Remove image backgrounds entirely in your browser using the Xenova/modnet model (Apache 2.0). Your image never leaves your device.

  • Free · no sign-up
  • Instant results
  • Privacy-friendly

AI-powered background removal that runs entirely in your browser. Drop a photo, get a transparent cut-out.

💡 Solid backgrounds (logos, screenshots, product shots) are removed instantly — no download, no waiting. Photos with complex backgrounds use an AI portrait model that downloads ~25 MB on first use (cached after that).
Drop a photo here or click to browse
JPG, PNG, WebP · up to 10 MB · full resolution

Other Image Tools

Your photo never leaves your device

This is the unusual part, and the reason to use it. Almost every background remover on the web uploads your image to a server, processes it there, and sends a cut-out back. This one does not. The MODNet segmentation model is downloaded into your browser and runs locally on your own hardware via transformers.js and WebAssembly.

There is no upload step, so there is no copy of your photo on anyone's server, no retention policy to read, and no question about whether your image becomes training data. You can confirm it yourself: load the page, disconnect from the internet once the model has downloaded, and the tool keeps working.

That property is what makes it appropriate for images you genuinely should not upload — ID photographs, pictures of children, medical images, unreleased product shots, or anything covered by a client NDA.

What to expect the first time

The trade-off for local processing is that the model has to reach you first. The initial load downloads several tens of megabytes, which takes a moment on a slow connection. Your browser caches it afterwards, so subsequent visits start immediately.

Processing speed then depends on your own machine rather than a server queue. A modern laptop handles a typical photo in a few seconds. An older phone will be slower, and very large images take longer, so downscaling an enormous photo before processing is worthwhile — the cut-out edge quality is set by the model, not by feeding it more megapixels than the subject needs.

What it handles well, and where it struggles

MODNet is a portrait matting model, so it is strongest on exactly the case most people need: a person or a well-defined object against a reasonably distinct background. Within that range it produces a clean alpha edge, including a decent attempt at hair.

It is weaker where the boundary itself is genuinely ambiguous. Fine flyaway hair against a busy background, transparent and semi-transparent material such as glass, veils or smoke, motion blur, and subjects wearing a colour very close to the wall behind them all produce a less certain edge. Low-contrast, poorly lit photographs are harder for the same reason.

The practical fix is nearly always at capture time rather than in software: more separation between subject and background, and even lighting on the subject, do more for the result than any post-processing.

Save as PNG or WebP, not JPEG

A cut-out is only useful if the transparency survives the save. JPEG has no alpha channel at all, so saving a cut-out as JPEG silently fills the transparent area with solid colour — usually black or white — and the work is lost.

Use PNG for maximum compatibility, or WebP for a substantially smaller file with transparency intact. Both preserve the alpha channel. If you are placing the cut-out onto a coloured background anyway, composite it first and then flatten to JPEG if file size demands it.

Licensing

The model used here is Xenova/modnet, released under Apache 2.0, which permits commercial use. This is worth checking whenever you use a background remover, because several popular alternatives — RMBG-1.4 among them — carry non-commercial licences that make them unsuitable for client or product work despite being free to download.

Frequently asked questions

Is my image uploaded anywhere? #
No. The model runs entirely in your browser through transformers.js and WebAssembly, so the image never leaves your device. You can verify it by going offline after the model has loaded — the tool still works.
Why is the first run slow? #
The model itself downloads to your browser on first use, which is a few tens of megabytes. Your browser caches it, so later visits start straight away. Processing speed afterwards depends on your own hardware rather than a server queue.
Why does the edge around hair look rough? #
Fine hair against a busy or similarly coloured background is the hardest case for any matting model, since the true boundary is genuinely ambiguous. More separation and contrast between subject and background at capture time improves the result more than any setting.
Which format should I save the result in? #
PNG or WebP. JPEG has no transparency support, so saving a cut-out as JPEG fills the transparent region with solid colour and destroys the effect. WebP gives a much smaller file than PNG with alpha preserved.
Can I use the output commercially? #
Yes as far as the model is concerned — Xenova/modnet is Apache 2.0 licensed, which permits commercial use. Worth checking elsewhere, since some popular background-removal models are licensed for non-commercial use only.
Does it work on objects, or only people? #
MODNet is trained primarily for portrait matting, so people are its strongest case. It handles well-defined single objects reasonably too, but struggles with transparent materials such as glass and with subjects that blend into the background.
Does it work on a phone? #
Yes, in any modern mobile browser, though processing is slower than on a laptop and the initial model download is more noticeable on mobile data. Downscaling a very large photo first speeds things up without hurting edge quality.