About AI Background Remover
AI Background Remover uses the IS-Net segmentation model to cut subjects cleanly out of any photo — people, products, pets, objects — and replace the background with transparency. Everything runs on-device through Hugging Face's transformers.js and WebAssembly, so your images never leave the browser. The model downloads once (~25 MB) on your first cutout and is cached locally; after that it works fully offline. No watermark, no resolution cap beyond your GPU/CPU, no sign-in.
- No uploads
- Browser-only
- Works offline
- 100% free
How it works
- 1
Drop your image
Pick a JPG, PNG or WebP. The original is rendered side-by-side so you can compare against the cutout.
- 2
Run the model
First time, the IS-Net weights are downloaded and cached. Each subsequent image runs straight from cache — usually a few seconds on a modern laptop.
- 3
Download a transparent PNG
Save the result as PNG with a true alpha channel, ready for product mockups, thumbnails or compositing.
Related tools
Browse allDouble image resolution with on-device AI super-resolution. Sharper edges, recovered detail. Model downloads once (~50MB), then offline.
Shrink JPEG, PNG, and WebP without visible quality loss.
Convert between PNG, JPG, and WebP in seconds.
Add arrows, boxes, highlights, blurs and text labels to any screenshot.
Frequently asked questions
Are my files uploaded to a server?
No. Every tool on SnapToolz runs entirely inside your browser using JavaScript and WebAssembly. Your file is read locally, processed in memory, and the result is offered as a download. Nothing is sent to a server — there isn't one to send to.
Which subjects does the model handle best?
IS-Net is a general-purpose dichotomous segmentation model — it does very well on people, products on neutral backgrounds, animals, vehicles and isolated objects. Hair wisps and fur edges are usually clean. It can struggle on transparent glass, motion-blurred edges, and subjects that blend into a similarly-coloured background.
Is the output AI-generated, and could it be imperfect?
Yes — the alpha mask is produced by a neural network and can have minor artefacts: a stray pixel along an edge, a small hole inside a complex subject, or a faint halo. For commercial use, eyeball the result and touch up in any image editor if needed. See our /disclaimer for details on AI output.
How big can my image be?
There's no hard cap, but very large images (above ~4000 px on the long edge) are downscaled internally before segmentation and then the mask is upscaled back. For pixel-perfect product shots, work at the original resolution and run a single subject at a time.
Does it work offline?
Yes. SnapToolz is a Progressive Web App. After your first visit, the app is cached on your device and every tool keeps working without an internet connection.
Is SnapToolz free?
Yes — every tool is 100% free with no sign-up, no watermark, no hidden tier. The whole platform is open source and we have no plan to gate features.