What happens when you upload
Your file leaves your device, crosses the network, and is written to storage on a server. It is processed there and the result is sent back. Several things follow from this that are not obvious:
The file exists in at least two places on the server side — the upload and the output — and often more, including temporary files and cache layers. Deletion policies vary from immediate to indefinite, and are stated in a privacy policy that most people do not read.
It may pass through a CDN, an object store and a queue on its way, each of which may retain a copy for some period. It may be backed up as part of routine infrastructure backups, which can extend retention well past the stated policy.
Staff may have access. Content may be scanned, for legitimate reasons like abuse detection and less legitimate ones. And a company's privacy policy binds only that company: if it is acquired or fails, the data is an asset that goes somewhere.
None of this means uploading is reckless. It means it is a decision with consequences, and the consequences scale with what is in the photo.
How browser processing works
Modern browsers can run real computation. Three capabilities make this practical for image work: WebAssembly compiles near-native code to run in the page, WebGL and WebGPU give access to the graphics hardware, and Canvas provides direct pixel manipulation.
A background remover built this way downloads the neural network model into the page — a few megabytes, once, then cached — and runs inference on your device using your GPU. The image is read from a local file input, processed in memory, and written back out as a download. There is no upload step because there is no server involved in the processing at all.
You can verify this rather than trusting it, which is the part worth knowing: open your browser's developer tools, go to the Network tab, and use the tool. If your image were being uploaded you would see the request. This is a claim that is checkable from outside, unlike a privacy policy.
Two more properties follow. There are no per-image costs, so limits and paywalls are unnecessary. And once the model is cached the tool works offline, because nothing needs the network.
Where the cloud genuinely wins
This is the honest part, and it is not close on some axes.
Raw capability is the big one. A server can run a model that is tens of gigabytes and needs a dedicated accelerator. A browser realistically works with models of a few tens of megabytes. For the very best quality on the hardest cases — complex hair against a busy background, fine transparency, semi-transparent fabric — a large server-side model still produces better results than anything that fits in a page.
Consistency is another. A server has known, fixed hardware. Browser performance depends on the visitor's device, which spans a decade of phones and laptops. The same tool can be instant on a recent machine and slow on an old one, and there is no way to fix that from our side.
Very large files are a real constraint. Browser tabs have memory limits, and a 100-megapixel image or a long video can exceed what a tab can hold. Servers do not have this problem.
Integration is the last one. If you need image processing inside an automated pipeline, an API on a server is the right architecture and a browser tool is not an option at all.
How to decide
The question worth asking is not which is better in the abstract but what is in the image.
For anything sensitive — identity documents, medical images, financial statements, photographs of children, unreleased work, anything under a confidentiality obligation — on-device processing removes an entire category of risk. There is no server-side copy to leak, subpoena, retain past its policy, or inherit in an acquisition. That is a structural difference, not a promise.
For ordinary images where you need the best possible result and the content is not sensitive, a good cloud service is a perfectly reasonable choice, and on the hardest cut-outs it will still beat a browser.
For bulk work, on-device usually wins on economics rather than privacy: no per-image cost and no rate limit means processing a few hundred product photos is just a matter of waiting.
Reading the claims
'Secure', 'private' and 'encrypted' are used loosely enough to be nearly meaningless in this space. Some specific things to look for:
- 'Encrypted in transit' means HTTPS, which every site has. It says nothing about what happens after arrival.
- 'We delete your files after an hour' is a policy, not a mechanism, and it depends entirely on the operator honouring it.
- 'Your files are never shared with third parties' does not mean they are not retained, scanned, or used for training.
- 'Processed locally' is checkable — open the Network tab and watch. If a request goes out with your file in it, the claim is false.
- A tool that works with your network connection off is not uploading anything, which is the most direct test available.