Background Remover
Cut the background out of a photo with a model that runs on your device.
Your files never leave your device — processing happens in your browser.
Drop an image here
or paste from your clipboard. Your file is processed on this device and is never uploaded.
How it works, and why that matters
Background removal uses a neural network trained to identify the salient subject of a photograph and produce a mask separating it from everything else. That mask becomes the image’s transparency, so the subject is kept and the background is cut away.
Almost every tool that offers this runs the model on a server, which means uploading your photograph. For product shots that may be fine. For photographs of people — which is what most background removal is actually used on — it means handing a picture of someone to a company whose retention policy you have not read.
Here the model is downloaded to your device and runs inside your browser. It is slower than a server GPU, and the model is smaller than the ones large services use, so the edges are not always as clean. That is the trade, and it is stated plainly rather than hidden.
Getting the best result
- Contrast helps most. A subject that differs clearly from its background in colour, brightness or focus produces a much cleaner mask than one that blends in.
- One subject works better than several. The model looks for the single most salient thing in the frame.
- Resize very large photos first. The model works at a fixed internal resolution regardless of your image size, so a 50-megapixel original costs a great deal of memory and time for no extra accuracy.
- Hair, fur and glass are hard. Semi-transparent and finely-detailed edges are where every segmentation model struggles, and a small one struggles more.
What you get back
A PNG at the original resolution with a transparent background. PNG because it is the format that everything understands and that keeps transparency — saving as JPG would fill the transparency back in with white and undo the work.
The file will usually be larger than the photograph you started with. That is expected: PNG is lossless, and your original was almost certainly a compressed JPG. If size matters, run the result through the compressor and choose WebP, which keeps transparency.
Questions
Is my photo uploaded to remove the background?
No, and this is the main reason to use this tool rather than a hosted one. The AI model runs on your own device. What gets downloaded is the model itself — about 5 MB of weights, once per device — which is code coming to you, not your photo going anywhere.
Why is there a download the first time?
Removing a background needs a trained segmentation model, and that model has to be on your device to run there. The alternative is sending your photo to a server that already has it, which is exactly what this site refuses to do. After the first use it is cached and the tool works offline.
Will there be a watermark?
No. The result is a full-resolution transparent PNG with nothing added. Watermarking the free result and charging to remove it is standard practice in this category and is precisely what this site exists to avoid.
Why is the edge rough around hair or fur?
Fine detail is the hardest thing for a segmentation model, and the one running here is small enough to download quickly, which trades some accuracy for speed. Clear separation between subject and background gives the best result — a well-lit portrait against a plain wall works far better than a busy scene.
Why is it slower than an online background remover?
Hosted tools run on server GPUs. This runs on your device’s CPU, which is genuinely slower for this kind of work. You are trading some seconds for the certainty that your photo was never sent anywhere. For a large image, resizing it first makes a substantial difference.
What does it work best on?
A single clear subject — a person, a product, an animal — against a background that differs from it in colour or focus. It struggles with transparent objects such as glass, with subjects the same colour as their background, and with several overlapping subjects.