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

说明与用途

ClearPix 通过在浏览器中下载 ONNX 模型并使用 WebGPU 进行推理，实现完全客户端的水印去除，确保隐私和高性能。

问题

Users are concerned that online watermark removers upload their photos to remote servers, risking privacy and data leakage.

结果

ClearPix achieves fully client‑side watermark removal with high performance (WebGPU reduces inpainting from ~3s to ~0.4s per region) and robust reliability through multi‑mirror fallback, self‑healing cache, and graceful degradation to WASM.

工作流

• 1. User opens ClearPix tool page in browser.
• 2. Browser downloads ONNX model weights from Cloudflare R2 via custom domain, verifies SHA‑256, and caches them in IndexedDB.
• 3. Browser loads ONNX Runtime Web (ort 1.30.x) via dynamic import, preferring WebGPU execution provider and falling back to WASM if necessary.
• 4. User uploads image or video locally; it is decoded, processed by the model, and re‑encoded within the browser.
• 5. No media bytes are sent over the network; only model weights and aggregate analytics events are transmitted.
• 6. Status bar displays real‑time download and inference status, confirming no uploads occur.

证据

When we enabled the WebGPU EP (ort 1.30+), our inpainting step went from 2100–3500 ms to roughly 400 ms per region in our benchmarks, and in October 2026 the HQ LaMa path finally joined it at 0.86 s per 512×512 tile.

来源

Models live in a Cloudflare R2 bucket behind a custom domain on our zone. This was a measured decision, not a default: direct r2.dev URLs gave us 75–110 KB/s, while the same‑zone custom domain delivered 2.3 MB/s — a 20–30× difference that turns a model download from a coffee break into a pause.

来源

The only bytes that travel are the models coming down to you, and a small set of aggregate analytics events that contractually cannot contain your media.

来源