Background Remover

About Background Remover

Background Remover uses the @imgly/background-removal AI library entirely in your browser to cut out image backgrounds without sending any data to a server. It applies a medium-accuracy segmentation model that downloads once (~40 MB) and is then cached locally for all future uses. The output is a transparent PNG you can drop into design tools, presentations, or social media. Because everything runs client-side, your images stay private — nothing is uploaded anywhere.

Why use Background Remover

• Runs entirely in your browser — no file uploads, no privacy risk.
• AI model is cached after first download, so subsequent uses are instant.
• Outputs a clean transparent PNG compatible with any design tool.
• Completely free with no watermarks or sign-up required.
• Works on photos, product shots, logos, and illustrations.
• Runs entirely in your browser using ONNX and WebAssembly — no file uploads, no privacy risk.

How to use Background Remover

1. Click the file input and select a JPG, PNG, or WebP image.
2. Click 'Remove Background'. On first use, wait for the ~40 MB AI model to download.
3. Watch the progress bar as the model processes your image.
4. Compare the before/after previews — the result shows transparency as a checkerboard.
5. Click 'Download PNG' to save the transparent output.
6. Click the file input and select a JPG, PNG, or WebP image up to roughly 10 megapixels.
7. Click 'Remove Background'. On first use, wait 20–60 seconds for the ~40 MB AI model to download and initialise.

When to use Background Remover

• Preparing product photos for e-commerce listings.
• Creating transparent logos or stickers from photos.
• Removing backgrounds from profile pictures for LinkedIn or ID cards.
• Isolating subjects before compositing in design projects.
• Preparing product photos for Shopify, Etsy, or Amazon listings that require white or transparent backgrounds.
• Creating transparent logos or stickers from photos for use on websites, slides, or merchandise.

Examples

Tips

• For sharp portraits, ensure the subject has clear separation from the background — flyaway hair against a similar-coloured wall is the hardest case for any matting model.
• If you plan to process many images, run the first one once to trigger the model download, then queue the rest — the cached model reuses the same WebAssembly instance.
• Resize huge images (above 4000 px on the long edge) before processing; the model runs internally at 1024 px so extra resolution increases time without improving quality.
• Save the transparent PNG losslessly — converting back to JPG will re-add a solid background and discard the alpha channel.
• If a photo has motion blur or heavy compression artefacts, run it through a denoise pass first; the matting model trusts edges, so noisy input produces noisy mattes.
• Use the result as a layer mask in your editor of choice if you only want partial transparency — the alpha channel preserves anti-aliasing for smooth blending.

Frequently Asked Questions

Explore the category

Glossary

ONNX

Open Neural Network Exchange — a portable model format used to ship the segmentation network from training frameworks like PyTorch into the browser via ONNX Runtime Web.

U^2-Net

A nested U-Net architecture used by the @imgly/background-removal library for salient-object segmentation; produces both a coarse mask and a fine alpha matte in one forward pass.

U-Net

An encoder-decoder convolutional network with skip connections, originally designed for biomedical segmentation and now ubiquitous in image-to-image tasks.

WebAssembly (Wasm)

A portable bytecode format that runs near-native machine code inside the browser; ONNX Runtime Web uses it to execute the AI model without a server.

IndexedDB

A persistent client-side database built into modern browsers; the tool stores the model weights here so the 40 MB download only happens once per browser profile.

Alpha matte

A grayscale mask describing per-pixel opacity — values between 0 (transparent) and 255 (opaque) preserve soft edges around hair and fur.

Salient object detection

The task of finding the most visually prominent object in an image; matting models build on this to produce per-pixel cutouts.