Image
Remove Background
AI-powered cutout — runs in your browser, never uploads your image.
Most background removers send your photo to a cloud server, process it, and stream back a result — which means your image travels the network and touches someone else's infrastructure. This tool does none of that. On first use, it downloads the BiRefNet AI model (~100 MB) from a public model hub and caches it permanently in your browser. Every removal after that runs entirely on your own hardware — no upload, no server, no account, no watermark. The output is a transparent PNG with the subject isolated from the background, ready to layer onto any design, slide, or product listing.
How to Use
- 1
Drop a PNG, JPEG, or WebP image onto the upload area, or click to browse.
- 2
On first use: the AI model downloads (~100 MB) and a progress bar shows the download. This happens once — the model is cached permanently in your browser after that.
- 3
Once the model is ready, background removal runs automatically. On WebGPU-capable hardware (Chrome, Edge) this takes a few seconds; the WASM fallback on other browsers takes slightly longer.
- 4
Preview the result on a checkerboard background to confirm the cutout, toggle between Before and After, then download the transparent PNG.
Frequently Asked Questions
Is my image ever uploaded to a server?
No. Everything runs in your browser. The AI model is downloaded once and cached; your image is processed locally and never leaves your device. You can verify this by opening your browser's Network tab — you will see no upload request while the tool runs.
Why does the first removal take longer?
The first run downloads the BiRefNet AI model (~100 MB) from a public model repository. A progress bar tracks the download. After that, the model is cached in your browser indefinitely — every subsequent removal starts immediately with no download.
What is the output format?
The output is always a transparent PNG. PNG is the only widely supported format that preserves the alpha channel (transparency) needed for a clean subject cutout. You can convert the PNG to WebP using the Image Converter tool if you need a smaller web-ready version.
Which AI model does this use?
BiRefNet (Bilateral Reference Network), a state-of-the-art MIT-licensed segmentation model. It runs via the Hugging Face Transformers.js library using ONNX Runtime in the browser — WebGPU for fast inference where available, WASM as the fallback. No proprietary or non-commercial models are used.