AI background removal that never sees your photo: a 4.6 MB model runs on your device, so it is private, free and unlimited.
PNG, JPG or WebP · works best with one clear subject · processed on your device, never uploaded
No size limit · files are processed on your device and never uploaded · how it works
First use downloads a 4.6 MB AI model once; after that it is cached and everything runs offline on your device.
Advanced settings (optional)
Cut-out options
New background
Transparent PNG/WebP — ready to place on any design.
Edge hardness
50
Low keeps soft, semi-transparent edges (hair, shadows); high cuts a crisp outline. Results re-render live.
Edge smoothing
1px
A small blur on the mask hides jagged outlines. 1–2 px suits most images.
Format
PNG keeps perfect quality and transparency.
Runs in your browser. Files are never uploaded — you can go offline after the page loads.
How to use the Background Remover
Drop one photo or a whole folder — PNG, JPG or WebP.
The AI model (4.6 MB) downloads once, then every image is processed on your device.
Tune Edge hardness and smoothing — the cut-out re-renders live.
Download transparent PNG/WebP files, or everything as a ZIP.
Easy to use
Drop files, pick options, download. Whole folders at once, no signup, no daily limit.
Best quality
The same engines the professionals use — SVGO, resvg, MozJPEG, libwebp, libavif — compiled to run in your browser.
Private & secure
Files are processed on your device and never uploaded. Close the tab and nothing remains anywhere.
Background removal without the upload
Every mainstream background remover works the same way: your photo is uploaded to a server, processed by a large model,
and handed back with a watermark, a signup wall or a per-image credit. This tool takes the opposite trade: a compact
salient-object model (u2netp, Apache-2.0) runs in your
browser through WebAssembly. Your image never leaves your device — which also means no file limits, no queue, no
watermark and no cost, and it keeps working offline once the model is cached.
What it is good at — and what it is not
A 4.6 MB model punches far above its weight on the jobs this site is built for: product photos, objects, logos,
screenshots of stickers and mockups, and portraits with reasonable contrast — the raw material of shop listings,
thumbnails and design work. It will not match a heavyweight cloud service on flyaway hair, fur, glass or subjects that
blend into a busy background. The Edge hardness slider decides how the model’s soft mask is cut
(soft edges for hair and shadows, a crisp outline for products), Edge smoothing hides jagged mask
edges, and both re-render instantly, so imperfect first results are usually a five-second fix. A “low confidence”
badge appears when the model could not find a clear subject.
After the cut-out
Export a transparent PNG for maximum compatibility or WebP for the same transparency at a much smaller size — or place
the subject straight onto white or a brand colour. From there: fit it into every platform frame with the
Social Media Image Resizer, scale it exactly with the
Image Resizer, or shrink it further with the Image Compressor.
Frequently asked questions
Are my photos uploaded to a server?
No. The AI model (u2netp, 4.6 MB) and its WebAssembly runtime are downloaded to your browser once and cached; after that every image is processed on your own device, even offline. That is why there is no file limit, no queue and no watermark — there is no server doing the work.
How good are the results?
Honest answer: very good for product shots, logos, and photos with one clear subject on a distinct background — the typical e-commerce or thumbnail job. A compact on-device model cannot match a heavyweight cloud service on fine hair, fur or busy scenes; the Edge hardness and smoothing sliders recover a lot in those cases, and everything re-renders live so tuning takes seconds.
What do the edge sliders do?
The AI produces a soft mask — each pixel is somewhere between background and subject. Edge hardness decides how that ramp is cut: low keeps semi-transparent edges (soft shadows, wispy hair), high forces a crisp outline. Edge smoothing blurs the mask outline by a pixel or two to hide jaggies. The result updates as you drag.
PNG or WebP for a transparent image?
PNG is universally accepted and pixel-perfect. WebP keeps the same transparency at a fraction of the file size and works in every current browser and design tool — choose it for the web, PNG for maximum compatibility (print shops, older software).
Can I put the cut-out on a new colour?
Yes — pick White or a custom colour under Cut-out options and the subject is composited onto it. For placing the cut-out into a specific frame (an Instagram post, a YouTube thumbnail), download the transparent PNG and drop it into the Social Media Image Resizer.
Most online converters upload your file to a server. InfyTool ships the actual encoders to your browser as WebAssembly instead, so processing happens inside your tab — faster for batches, and nothing to delete afterwards.
No upload, ever. Turn off Wi-Fi after the page loads and every tool keeps working.
No limits. No file-size cap, no queue, no watermark, no account.
Runs on your CPU. Batches use several cores in parallel; nothing is throttled.