The short answer
This tool's detection catalog, removal engine, browser workspace, and CLI are public, MIT-licensed source code you can read directly in the open-source repository, not a closed service you have to take on faith. Processing runs entirely in your browser, so there is nothing uploaded to trust in the first place, and the code that proves that is the same code you can read, fork, and run yourself.
What is actually in the repository
The public repository is not a marketing summary of the product; it is the literal code that runs when a file is processed. It includes the watermark size and margin catalog, the detection and removal engine built on top of it, the browser workspace that powers this site's eraser and video tools, the command-line tool and npm package built on the same engine, and the automated tests that check all of it against real sample fixtures. Nothing about how a file gets processed lives outside what is checked into that repository.
Why this matters more for a watermark remover specifically
A watermark remover sits in an unusually sensitive spot: it touches images people may not have full rights to redistribute, and it makes a specific technical claim about what it will and will not alter. A closed tool asking you to trust both of those things at once is asking for a lot. An open, inspectable codebase is a stronger answer to "can I trust this" than any amount of copy on a landing page, because it lets a technically capable visitor check the documented scope limits against the actual detection logic instead of a page's word choice.
How to verify the claims yourself, not just read them
Every honesty claim on this site is only as good as how easily it can be checked. Concretely, for this project:
- No upload. Open your browser's network panel while using the eraser or video tool. Your source file never appears as an outgoing request body; only this site's own static assets and, for video, the local model files load.
- No account, no server-side processing. There is no sign-in flow to inspect and no server-side endpoint that receives your media, because there is no server in that path at all.
- No telemetry. The repository has no analytics, session-replay, or tracking dependencies in its processing path. Reading
package.jsonand the source tree is enough to confirm that directly, rather than relying on a privacy policy's wording. - The detection math itself. The size catalog page states specific pixel dimensions, watermark sizes, and margins; every one of those numbers is generated from a single source file in the repository, so it can be checked against the running code rather than against this page's prose.
The delivery surfaces, and where each one's code lives
- Web app — this site's eraser and video tools, running the engine directly in your browser tab.
- Chrome extension — installable from the Chrome Web Store listing, for processing images without leaving the page you found them on.
- Userscript — a Tampermonkey-compatible script, published as a direct release download, for running the same processing logic inline on supported pages.
- CLI and npm package — published as @pilio/gemini-watermark-remover on npm, and covered in more depth on the batch processing page, for scripting removal across many files at once outside a browser tab.
What open source does not automatically mean
Publishing the source is a necessary condition for auditability, not a substitute for it. A repository nobody reads is not meaningfully more trustworthy than a closed one just because its license file says so. This page is written for the visitor who actually intends to look, which is why it points at specific files and claims worth checking rather than only the repository root. The same standard applies everywhere else on this site: a claim here is only as good as how easily it can be checked, and open source is what makes that checking possible in the first place, not a substitute for someone actually doing it.
License and reuse
The repository is published under the MIT License, a permissive license that allows reuse, modification, and self-hosting with attribution, without a copyleft obligation to publish your own changes. Anyone considering a fork, a self-hosted deployment, or a derivative tool should read the published license file directly rather than assume a default, since a summary paragraph is exactly the kind of claim this page argues you should not take on faith. The npm package and CLI expose the same engine for local or self-hosted use if the hosted web app is not what you need.
Last updated 2026-07-31