Remove Watermark Without Uploading: How We Built an In-Browser AI Editor

Описание и применение

ClearPix реализует удаление водяных знаков полностью в браузере, загружая модели ONNX только один раз и кэшируя их в IndexedDB. Весь процесс инференса выполняется на WebGPU (с резервом WASM), гарантируя, что пользовательские изображения никогда не покидают устройство.

Задача

Users are uncomfortable uploading sensitive photos to online watermark removers that process images on remote servers, raising privacy concerns.

Результат

ClearPix achieves privacy‑first watermark removal by moving all inference to the client, reducing first‑visit latency through prefetching, and ensuring structural privacy via client‑side caching and no‑upload guarantees.

Рабочий процесс

• Open ClearPix tool page
• Browser downloads ONNX model weights from R2 CDN (with mirrors and fallback)
• Model weights are cached in IndexedDB for future use
• ONNX Runtime Web creates a session preferring WebGPU, falling back to WASM if needed
• User uploads image/video locally
• Image/video is decoded, processed, and re-encoded entirely in the browser
• Result is displayed to the user
• Optional aggregate analytics events are sent (no media data)

Доказательства

Your image goes up, gets processed on someone else's GPU, and comes back down. Whether the privacy policy says the file is deleted afterward, you have no way to verify it, and neither do we.

Источник

When you open a tool page, your browser downloads ONNX model weights (once, then cached forever), loads them into ONNX Runtime Web, and runs inference on WebGPU where available with a WASM fallback.

Источник