Help & privacy

How it works

1. Upload JPG, PNG or WebP images (drag & drop, choose, or paste).
2. In Cutout, choose a model and press Remove background. The first run downloads the model once.
3. Refine with the erase/restore brushes, feathering and expand/contract.
4. Pick a background, adjust colours, resize, and add shadows.
5. Use Export for PNG/JPG/WebP, or Compress for size-optimised files. Batch results download as a ZIP.

Where your images are processed

Everything — decoding, background removal, editing and encoding — runs on your device in the browser. Your images are not uploaded to our server or any third-party service. The only network downloads are the application itself and the model weights, fetched from Hugging Face (or the site’s own model host if configured) and cached by your browser. If you start from an upload page on clippingworld.com, the image passes through our server once so it can reach the editor: it is held for at most 10 minutes and deleted the moment the editor picks it up.

Storage & retention

Projects (original image, masks and settings) are autosaved in this browser’s IndexedDB so you can recover them after a refresh. They stay until you delete them (trash icon in the image tray) or clear site data. They are not synced across devices or browsers. The server keeps no copies of your images, so there is nothing server-side to delete.

Metadata

Exports are re-encoded by the browser, which writes no EXIF, GPS/location or camera metadata. Orientation from the original EXIF is applied to the pixels on import.

Models & licences

• General (IS-Net) — BritishWerewolf/IS-Net, Apache-2.0 licence. People, products, animals and objects. ~179 MB one-time download, cached by your browser.
• Fast portrait (MODNet) — Xenova/modnet, Apache-2.0 licence. Optimised for people/portraits only. ~7–13 MB download.
• High detail (BiRefNet-lite, WebGPU) — onnx-community/BiRefNet_lite-ONNX, MIT licence. Finer edges; needs a WebGPU-capable browser/GPU (≈115 MB fp16). Too memory-hungry for WebAssembly.
• Inference runtime: transformers.js / ONNX Runtime Web (Apache-2.0), WebGPU when available, otherwise WebAssembly.

Limits

• Upload: 25 MB and 40 MP per file.
• Output: up to 16.7 MP (browser canvas limits, especially on iPhone/iPad).
• Enlarging uses interpolation — it does not add detail and is not “AI upscaling”.
• Automatic cutouts are not perfect on every photo; fine hair, glass and low-contrast edges may need brush work.
• Unsupported: animated images, GIF, HEIC/HEIF, AVIF input, SVG.

Keyboard shortcuts

Open the editor