Remove image backgrounds in your browser with AI
Background removal used to require Photoshop and a steady hand on the magic-wand tool. Then it became a server-side AI service — fast, but you uploaded your photo to a stranger's machine and often paid per image. With recent advances in WebAssembly and ONNX Runtime Web, the same neural network that powers commercial background removers can now run directly in your browser. That's what this tool does.
How AI background removal works
Behind the scenes, the tool uses a deep learning model (a salient-object segmentation network, similar to U²-Net or BiRefNet) that's been trained on millions of images to predict, for every pixel, whether it belongs to the subject or the background. The model runs locally in your browser via ONNX Runtime Web. Your image is converted to a tensor, fed through the network, and the output is a binary mask we use to cut out the background. The whole thing takes 3–10 seconds on a modern laptop, longer on phones.
What kinds of photos work best
- Portraits — single person, even backgrounds, good lighting. Usually crisp results.
- Product photos — for e-commerce listings, where you need a clean white-background product shot from a casual photo. This is the killer use case.
- Pet photos — works surprisingly well on fur edges thanks to alpha-matting in the model.
- Object photography — single isolated objects (shoes, jewelry, food items) for catalogs or social media.
Where it struggles
Glass, transparent objects, fine hair against a similar-toned background, group photos with overlapping people, and motion-blurred subjects — the model can produce edge artifacts. For perfect-quality work, you'll still need Photoshop. For 90% of practical use cases, this tool gives you a 30-second result that previously took 30 minutes.
Compared to remove.bg and other paid services
remove.bg charges $9/month for basic and pricier tiers for higher resolution. Canva, Photoroom, and similar tools all upload your photo. Toolvana does not — the model runs in your browser, the bytes stay on your device, and we don't cap the resolution or the number of uses.
FAQ
Is the model really running in my browser?
Yes. The first time you use the tool, your browser downloads the model file (~40 MB) and caches it. After that, even offline, the tool works. Open DevTools → Network and you'll see zero outbound requests for your image.
Why is the first run so slow?
The model file has to download. Once cached, subsequent runs are 3–10 seconds depending on image size and device.
What image formats are supported?
JPG, PNG, WebP. Output is PNG with transparency (or JPG with the background color you pick).
Is my image really not uploaded?
Confirmed — open the Network tab in DevTools while you process. You'll see model file downloads but never your image leaving.
What's the max image size?
Bound by your device's RAM. Most desktops handle 4K images fine. The model internally works at a fixed resolution and upscales the mask back.
Can I use this for commercial work?
Yes. Outputs are yours to use however you want. Toolvana doesn't claim any rights.