About this tool

Why it is free

Most online image tools are free in the sense that you pay with an account, a watermark, a daily limit, or your photo. Those trade-offs exist because running an image model on a server costs real money per image, so the operator has to recover it somehow.

This tool removes that cost entirely by not using a server for the work. The model runs on your device, using the hardware you already own. There is no per-image cost for us to recover, which is why there is no account, no quota, no credit system, and no watermark: there is nothing to meter and nothing to sell back to you.

How it works, technically

When you add an image, it is decoded in your browser and passed to a background-removal neural network. That inference runs in a dedicated web worker so the interface stays responsive. The tool prefers WebGPU, which executes the model on your graphics chip, and falls back to WebAssembly, which runs it on your CPU, on devices or browsers where WebGPU is not available. The performance difference between the two paths is significant, which is why the tool reports which engine it used.

The result is composited onto a transparent or solid background in the browser and offered to you as a download — PNG for transparency, or JPG and WebP for flattened images. The image itself never crosses the network.

The model, and its licence

The shipped model is IS-Net, a general-purpose image segmentation network, shipped here in a quantized form (~44 MB download) whose weights are derived from the Apache-2.0 licensed DIS release. That licence permits commercial use. This matters: several popular background-removal models — including BRIA's RMBG-1.4 — are released for non-commercial use only, which is why they are deliberately not used here. A free tool that quietly ships a non-commercial model is a legal problem waiting to be discovered.

IS-Net was chosen because it generalises: people and portraits, but also products, animals, logos and mixed scenes, where the previous portrait-specialist model visibly struggled. It reasons about the image at 1024×1024, so the first cutout on a device can take a little longer than before. Very fine hair, fur, and translucent objects remain genuinely difficult for this class of model, and the tool says so on the home page rather than pretending otherwise.

What it is not

This is a single-purpose utility, not a design suite. It has no layers, no templates, no text tools, no background image library, and no video support. That restraint is intentional: the goal is to do one job quickly, on your own hardware, without asking you to sign up.

Who it is for

Small marketplace sellers who need clean product photos at exact dimensions, freelancers who need a quick cutout without uploading a client's confidential asset, office workers making slides, and anyone who needs a transparent image once and does not want to create an account to get it.