One-Click Background Remover
Cut out a person, a pet or a common object automatically and export a transparent PNG. Runs locally with MediaPipe - no upload and no sign-up.
Drop your image file here
or click to browse files
Pick an image and the tool automatically separates the subject from its background, then exports a transparent PNG you can drop onto any colour or design. Three models are available: a compact person model (0.24 MB), a 21-class model for common objects such as cats, dogs, cars and bottles (2.65 MB), and a finer person model that keeps hair and clothing detail (15.61 MB).
You can preview the result side by side with the original, tighten the mask edge to remove the usual halo of leftover background pixels, and place the cut-out on a transparent, white, blue, red, dark or green background - which is exactly what shop listings, ID photo rules and app icons ask for. Everything runs on your own device: the first run downloads the segmentation runtime (about 4 MB) and the chosen model, both cached afterwards, and the picture itself is never uploaded.
How to cut out a subject and remove the background
Add a photo with a clear subject
Choose a model and refine the edge
Pick a background and download the PNG
FAQ
How much is downloaded the first time?
About 4 MB for the MediaPipe runtime plus the model you choose: 0.24 MB for the person model, 2.65 MB for common objects, or 15.61 MB for the fine person model. Your browser caches both, so later visits download nothing again.
Which model should I pick?
Person / selfie for people and portraits - it is small and fast. Common objects when the subject is a pet, a car, a bottle or another ordinary object. Person + hair + clothes when frizzy hair or a detailed outfit has to keep clean edges.
Why is there a halo or leftover edge?
Segmentation masks are approximate at the boundary. Raise the edge refinement slider to erode the mask slightly, which removes the thin ring of background pixels that often survives.
Can it cut out anything?
No. The models know people, hair, clothing and 21 common object classes. An unusual object or a very cluttered scene will need manual masking in an image editor - we would rather say that than pretend it works everywhere.
Does it work on a phone?
Yes, but the model runs on your device, so a mid-range phone takes a few seconds per image and the fine 15.61 MB model may be slow. The person model is the quickest choice on mobile.
Is my photo uploaded?
No. The model runs in your browser with WebAssembly, and the image never leaves your device - which matters for portraits and identity documents.
At a glance
| Input | image |
|---|---|
| Multiple files | No |
| Where it runs | In your browser - files are never uploaded |
| Works offline | Yes — once the page has loaded |
| Price | Free, no sign-up, no watermark |
How much quality you actually lose
Lossy compression throws away detail your eye is least likely to notice - first fine texture, then edges. In practice most photographs look identical to the original down to roughly 75-80% quality, and differences only become obvious around faces, hair, foliage, fabric patterns and text. That is why a portrait and a screenshot need different settings: a face at 60% shows blotchy skin where a landscape at 60% still looks fine. Resizing matters more than quality for perceived sharpness: halving the pixel dimensions usually saves far more than dropping quality from 90 to 60, and the result looks better. If you need a specific file size, reduce dimensions first, then quality, and stop as soon as you hit the target rather than going lower than necessary.
Choosing between JPG, PNG, WebP and AVIF
These formats solve different problems, and picking the wrong one costs you either quality or file size. JPG is the safest choice for photographs and for anything you need to send to someone else, because every device and program opens it - but it is lossy and has no transparency. PNG is lossless and supports transparency, which makes it right for logos, screenshots, diagrams and anything with sharp edges or text, at the cost of a much larger file. WebP typically beats JPG by 25-35% at the same visible quality and supports transparency, but a few older programs still cannot open it. AVIF compresses even further, especially at low quality settings, but encoding is slower and support is the newest of the four. A practical rule: JPG to send, PNG to edit, WebP for your own website, AVIF when every kilobyte counts and you have checked that your audience can open it.