About WatermarkOut

🎯 Why this exists

Watermark removers typically run on someone else's servers: you upload a private photo, pay a monthly subscription, and often get a result with their watermark on it unless you upgrade. That's broken.

WatermarkOut runs entirely in your browser. Your file never leaves the tab. No subscriptions, no accounts, no cloud — and because the source is MIT-licensed, you can check that claim rather than trust it. Apart from the app's own files and a one-time 29 MB model download on your first visit, nothing goes over the network.

⚙️ How it works (the boring honest version)

1. You drop an image in. The file becomes an in-memory ImageData object in your browser. It never leaves.
2. You paint the watermark. A canvas overlay tracks where you brush — that becomes the inpainting mask.
3. An AI model runs in WebAssembly. The model (MI-GAN, 29 MB, downloaded once and cached) runs inside your browser tab via ONNX Runtime Web — on your GPU if your browser supports WebGPU, otherwise on your CPU.
4. You download the result. Same browser, same tab, same machine. No server saw a thing.

Open your browser's Network tab and check for yourself: once the app and the model are cached, removing a watermark makes no network requests at all.

🧰 Other free tools by DiceCodes

🔧 Built with

• Vanilla HTML/CSS/JS — no build step, no framework, no node_modules
• MI-GAN — mobile/browser-optimized inpainting (Picsart Research, ICCV 2023)
• ONNX Runtime Web — WebGPU + WASM AI inference in the browser
• Canvas 2D — masking, cropping, and compositing (no image libraries)
• IndexedDB + Service Worker — offline support
• GitHub Pages + Hugging Face — free hosting forever

Source: github.com/IamRamgarhia/free-ai-watermark-remover · License: MIT · Audit anything. Fork anything.