PNG
Lossless, supports full transparency. Best choice when you need a clean transparent cut-out for design work.
Input · Output
Remove image backgrounds automatically using AI directly inside your browser. Create transparent PNG images, edit backgrounds and download your results — without uploading your photographs to any server.
Drag a file into the panel below, browse your device, paste from the clipboard, or capture a photo directly with your camera where supported. Processing happens locally — nothing is uploaded to a server.
Drop your image here
PNG · JPG · JPEG · WEBP — up to 20 MB
Your image is processed on your device and never uploaded.
Preparing…
Initialising
BG Remover is a complete background-removal workspace. It combines a trained segmentation model, a manual refinement editor, and multi-format export — all inside a single static page.
A custom-trained ONNX model identifies the main subject and separates it from the background in a single pass. No manual tracing required.
Every pixel stays on your machine. There is no server, no upload, no account and no tracking. The AI model is downloaded once and cached.
The AI processes a normalised tensor, but the alpha mask is resampled back to the original image dimensions, so your export keeps every pixel.
Fine-tune edges with soft brushes, hardness control, and pressure support for pen and touch input. Undo and redo are unlimited within your session.
Keep transparency, choose a solid colour, apply a gradient, or drop in a custom background image — with fit, fill and positioning controls.
Export transparency-preserving PNG or WebP, or flatten onto a chosen background as JPEG. Choose Standard, High or Maximum quality.
BG Remover detects WebGPU support and switches automatically, falling back to WebAssembly when hardware acceleration is not available.
Install BG Remover to your home screen. After the first successful load, the AI model and application shell are cached so subsequent runs work offline.
BG Remover is not a wrapper around a cloud API. Every step below — decoding, inference, refinement and export — happens inside your browser.
When you select an image, the browser decodes it into an
ImageBitmap and validates format, dimensions and file size.
Nothing has left your device at this point.
The image is resized to the model input size (typically 512×512) and
normalised into an RGB float tensor. The mean and standard deviation used here
are documented in js/ai/model-config.js.
ONNX Runtime Web executes the trained segmentation model using WebGPU when available, or WebAssembly otherwise. The model returns a foreground probability mask.
The mask is upsampled to original resolution with bilinear filtering, edge anti-aliasing is applied, and small noise components are removed before the alpha channel is composited.
The alpha mask is applied to the original image, producing a transparent PNG in memory. You can then refine edges, swap the background, or download immediately.
BG Remover works with the most common photographic formats. Choose a format based on how you plan to use the result.
Lossless, supports full transparency. Best choice when you need a clean transparent cut-out for design work.
Input · Output
Excellent for photographs and smaller file sizes. Does not support transparency, so the background must be replaced before export.
Input · Output
Modern format that supports both transparency and lossy compression. Great balance between quality and file size for the web.
Input · Output
Many online background removers upload your photos to a server. BG Remover takes the opposite approach.
BG Remover ships with a custom-trained image-segmentation model exported to ONNX. There is no cloud API involved — the model and the runtime are served as static files, and inference happens inside ONNX Runtime Web.
The model is a compact encoder-decoder network trained to predict a foreground probability mask from a normalised RGB image. Because inference is local, the only cost to run it is your device's time.
Yes. BG Remover does not charge, does not require an account, and does not call a paid API. Everything you see — the model, the runtime, the editor — is served as static files. Your only cost is the bandwidth to download the model the first time.
No. The image is decoded, resized, normalised, passed through the ONNX model, and exported — all inside the browser. There is no upload step anywhere in the pipeline.
You can upload PNG, JPG, JPEG and WEBP. You can export PNG with transparency, WEBP, or JPEG flattened onto a background.
After your first successful load, the service worker caches the application shell and the model. If your browser supports the Cache API and IndexedDB, subsequent uses of BG Remover can run completely offline.
Yes. The AI model runs at a fixed input size, but the resulting mask is upsampled and applied to the original image. The exported file keeps the source's resolution unless you choose a smaller preset.
The AI model is a trained neural network. It must be downloaded once so that your browser can run it. After the first download, the model is cached and reused on subsequent visits.
BG Remover automatically falls back to WebAssembly, which works on every modern browser. WebGPU is faster when available, but it is not required.
No sign-up. No uploads. Just drop an image and download a transparent PNG.