- Your images are never uploaded. There is no server here that could receive one.
- There are no accounts, so there is nothing about you to hold.
- We set no cookies and run no analytics, tag manager or tracking pixels.
- The only things your browser downloads from us are the pages, the fonts, the runtime and two model files.
- Anything the tool keeps — the models, your theme, a file in transit — stays on your device.
Your images are never uploaded
When you open a photo here, the file is read from your disk straight into a canvas in the browser tab. The segmentation model runs over those pixels in your own machine’s memory, through onnxruntime-web, compiled to WebAssembly. Compositing, resizing and encoding happen in the same tab, and the finished file is handed to your browser’s download manager. At no point in that sequence is there a request carrying your picture.
This is architectural, not a promise about our conduct. The site is a set of static files — HTML, CSS, JavaScript, two model files and a WebAssembly runtime — served from a CDN. There is no application server, no database and no storage bucket behind it. A hosted background remover can promise to delete your uploads afterwards; we are not in a position to make that promise, or to break it, because the images are never in our hands to begin with.
Nor is it something we could quietly change. Sending your image anywhere would mean shipping new JavaScript and standing up something to receive it, and the request would appear in your browser’s network panel — on a tool whose entire argument is that the panel stays empty. Open developer tools, switch to Network, and process a picture: after the one-time model download there are no further requests, which is also why the editor keeps working with your network disconnected.
The same is true of anything else you open in the editor, including a background image of your own. It is loaded from your disk into the same tab, and it is as private as the photo you put in front of it.
There is a longer walk-through of the mechanism, with the actual request list, in how on-device processing works.
What your browser fetches from us
Traffic goes one way: you download the tool, and nothing goes back. These are the files, by name.
| File | What it is | Size |
|---|---|---|
| HTML, CSS and JavaScript | The pages themselves, and the script that runs the editor. | tens of kB |
| /_astro/fonts/*.woff2 | Geist and Geist Mono, served from this origin rather than from Google Fonts. | 52 kB |
| /ort/ort-wasm-simd-threaded.wasm | ONNX Runtime — the WebAssembly engine that executes the model in your tab. | 13.3 MB |
| /ort/ort-wasm-simd-threaded.mjs | The runtime’s worker glue, needed for multi-threaded inference. | 24 kB |
| /models/u2netp.onnx | The Fast segmentation model (U²-Net small · 4.4 MB). Loads on open, so the first cutout is immediate. | 4.4 MB |
| /models/isnet-general-use-q8.onnx.001 – .003 | The Quality segmentation model (IS-Net int8 · 42 MB), split into three parts because our host rejects single assets over 25 MiB. | 42.2 MB |
All of it comes from this domain. Nothing is pulled from a third-party CDN, a font service or anyone else’s server, which is the thing the privacy claim actually rests on — a third-party script would see your visit even if it never saw your photo.
After the first load the runtime and both models sit in your browser’s cache storage and are not requested again.
Those are ordinary requests for public files, and like any web host, ours records them: an IP address, a timestamp, a user agent, and which file was asked for. That is how the web works and we have not found a way to serve a file without it. What those logs never contain is your image, because your image is never a request.
What is stored on your device
Three browser stores are used, all of them local to your machine and none of them readable by us. Browser storage is not a cookie: it is not attached to requests and it is not sent anywhere.
| Store | Holds | Cleared |
|---|---|---|
| Cache Storagethechangebg-models-v2 | The model weights and the runtime, so later visits need no network at all. | When you clear site data. Weights from an older model generation are deleted automatically the next time the editor loads. |
| IndexedDBthechangebg › handoff › pending-files | A file you dropped on the home page, waiting for the editor to pick it up across the navigation. | The editor deletes the entry as it reads it, so a refresh does not re-import the same photo. |
| localStoragethechangebg:theme | Whether you chose the light or the dark theme. | When you clear site data. |
The IndexedDB entry is the one worth explaining. Dropping a photo on the home page has to get that file to the editor, which is a different page — so the file is written to IndexedDB, the browser navigates, and the editor reads it back and deletes it. IndexedDB is per-origin storage on your own disk; the file never crosses the network, it just crosses a page load.
You can wipe all three from your browser’s site-data settings at any time. The only consequence is that the models download again on your next visit.
Analytics
There is no analytics on this site. Not Google Analytics, not a tag manager, not Cloudflare Web Analytics, not a pixel, not session recording, not an A/B testing script. The site ships no third-party JavaScript of any kind, so nothing is counting your visit but the web server’s own request log.
We would rather say that plainly than describe a “privacy-friendly, cookieless” setup we have not actually turned on. If that changes, the measurement we would add is cookieless and aggregate — no personal data, no cross-site identifier — and this section will describe it, with the date at the top of the page changed, before it ships.
What this site does not have
- No accounts
- There is no sign-up, no login, no email capture and no waiting list that collects an address. Nothing on this site asks who you are, which means there is no profile, no usage history and no password to lose.
- No cookies
- Nothing in our code reads or writes a cookie. Your theme choice and a file in transit use browser storage, described above, which is a different mechanism and is never transmitted.
- No third-party embeds
- No YouTube players, no chat widgets, no social buttons, no hosted fonts. Every asset is same-origin. The site is also served cross-origin isolated — the
COOPandCOEPheaders the model needs for multi-threaded inference — which means a cross-origin resource cannot simply be dropped in without deliberately arranging for it. - No advertising
- No ad network, no retargeting, no data brokers, and no sale or sharing of personal data — there is none here to sell.
- A few outbound links, and nothing hidden in them
- Some pages link to remove.bg, Canva and a news report about the closure, so the claims can be checked at the source. Following one is an ordinary navigation to a site with its own privacy policy. The links carry no tracking parameters.
Your data rights
Rights to access, correct, export or delete personal data attach to personal data that someone holds. We hold none: there is no account to export, no profile to correct and no record of your images, because we never had them. The material the tool does keep is on your device and under your control, and clearing your browser’s site data removes it.
If you believe something here is collecting more than this page describes, we would genuinely like to know — that would be a bug in the strongest claim we make. Write to us and we will look at it.
Changes, and how to reach us
If this policy changes, the date at the top of the page changes with it. A change that made images leave your device would be a different product, and it would be described here in plain words rather than absorbed into a paragraph about “processing partners”.
Privacy questions go to [email protected], or through the contact page. The terms of use cover the rest of the relationship: what you may do with the output, and what we do not warrant.