Magic Cutout Pro
Tap any object to cut it out — surgical, multi-object, on-device.
Runs entirely in your browser — nothing you enter is uploaded, logged, or tracked.
A surgical, interactive cutout tool where you tap exactly what to keep — and a real Segment-Anything neural network (SlimSAM) running 100% on your device traces a clean edge around that object. Where an auto background remover guesses one foreground subject, Magic Cutout Pro lets you point at the precise object under your finger, refine the selection with add (＋) and remove (−) taps, and stack several objects into a single cutout. Then export a transparent PNG sticker, composite onto a solid colour, blur or replaced background, or add a stickerize outline. The AI model downloads once (~40 MB) and then works offline; your photos never leave your device, and every export is re-encoded so GPS and camera metadata are left behind.
Export
Re-encoded on your device — no GPS or camera metadata, no watermark.
Frequently asked questions
Is my photo uploaded anywhere?
No. The segmentation AI and every effect run inside your browser on your own device — no servers, no upload, no analytics. After the one-time model download you can disconnect from the internet and it still works. Cloud cutout tools (remove.bg, Photoroom, Cleanup.pictures) send your photo to their servers; this one never does.
How is this different from the AI Background Remover?
The Background Remover automatically guesses the one main subject. Magic Cutout Pro is interactive and surgical: you tap exactly the object you want, refine it by tapping more spots to add or a “Remove” tap to trim, and you can keep several separate objects in one cutout — things an auto-remover simply can’t do. Use the auto remover for a quick one-subject cut; use this when you need to choose precisely.
How does an AI run in my browser with nothing uploaded?
It uses SlimSAM — a compact version of Meta’s Segment Anything model — compiled to run on your device via WebAssembly. The first time you open an image it spends a few seconds “reading” the photo once (a progress bar shows it); after that, each tap returns a precise mask almost instantly. The model downloads once (~40 MB), caches, and from then on everything is computed locally and offline.
Can I cut out more than one object?
Yes. Tap the first object, press “Add object”, tap the next, and so on — the cutout is the union of everything you picked. You can also switch to “Remove selected” to erase the tapped objects instead of keeping them, and export each object as its own transparent PNG layer in a .zip.
Will the cutout have real transparency?
Yes — export PNG or WebP and the background is a genuine alpha channel, not a white box, so you can drop the cutout onto anything. You can also composite onto a solid colour, a blurred version of the original, or a new background image and export a JPEG.
How good is it, and what works best?
It’s excellent on objects with a defined edge — people, products, animals, furniture, signs. Fine hair, fur, glass and chain-link stay hard for any compact on-device model; the edge-refine sliders (feather, crispness, grow/shrink) help but don’t fully solve them. The mask is computed at a modest resolution and upscaled, so it’s a clean social/sticker-quality matte, not a 4K rotoscope. It trades a little of a giant cloud model’s peak quality for total privacy and offline use.
Does it remove EXIF / GPS metadata?
Always. Every export is re-encoded from raw pixels, so the original’s GPS location, camera model and EXIF tags can’t carry over — there’s nothing to toggle.