Cutout Foreverruns on your device ☕Tip

How it works

  1. The page and the AI model download to your browser. On browsers with WebGPU the model is BiRefNet-lite (MIT licence, about 115 MB). Elsewhere it is ORMBG (Apache-2.0, about 44 MB), which fits in the CPU engine's memory. Either is fetched once from the Hugging Face CDN and cached.
  2. Your image is read locally. It is resized and fed to the model with ONNX Runtime Web on your GPU (WebGPU) or CPU (WebAssembly).
  3. The model returns a mask, which is applied as transparency at your export size. The PNG is created in your browser and saved straight to your device.

At no step is your image sent anywhere. The only requests the page makes are for its own code, the model files (Hugging Face), the ONNX Runtime engine (jsDelivr) and (if you use Pro) your sign-in status. You can verify this in your browser's developer tools (Network tab).

Pro limits are checked in your browser. Your account stores your email and subscription status only.