Privacy information
Initial website information · 31 October 2026
Zhatpat AI is being developed. This page describes the intended privacy approach for the website at this early stage. It is not a claim that every planned tool is already available.
Files and images
The Image to Text OCR beta processes the selected image in your browser using Tesseract.js. The image is not uploaded to a Zhatpat AI application server. The OCR JavaScript engine and the selected language data are downloaded from a third-party content delivery network when the tool is used. The Video Watermark Remover beta processes video locally in the browser and exports a WebM file. It offers manual selection, heuristic corner-logo candidate scanning and confidence-scored template tracking, then masks the selected region; it does not restore the original pixels behind a watermark. The tool does not upload video to a Zhatpat application server. The Background Remover beta uses the open-source MODNet portrait-matting model in the browser. Model files are downloaded from Hugging Face and cached by Transformers.js where supported; uploaded photos are not sent to a Zhatpat application server. Input files are limited to 8 MB and 24 megapixels, with processing downscaled to about 6 megapixels for memory safety. The Object Eraser beta uses OpenCV.js and classical Telea inpainting locally in your browser. The OpenCV.js library is loaded from the official OpenCV documentation CDN; the selected image is not uploaded to a Zhatpat application server. The AI Object Eraser beta uses the MI-GAN ONNX model with ONNX Runtime Web. Its model (about 28 MB) is downloaded from Hugging Face, SHA-256 verified, and cached in browser storage when supported; transient model-host throttling is retried a limited number of times; the selected image is processed locally and is not sent to a Zhatpat application server. Input is limited to 1 megapixel for device-memory safety. The Photo Restoration beta uses the Real-ESRGAN ONNX model with ONNX Runtime Web. The model is SHA-256 verified and cached in browser storage when supported; the runtime is downloaded from jsDelivr; the selected photo is processed locally in the browser and is not sent to a Zhatpat application server. The first use downloads the model (about 4.9 MB), which may be cached by the browser. This beta is limited to images up to 1 megapixel and may use substantial device memory. The AI Image Upscaler beta uses the Apache-2.0 Clark Swin2SR Lightweight ×2 ONNX model with ONNX Runtime Web. Its compressed model (about 1.7 MB) is SHA-256 verified and cached in browser storage when supported; transient model-host throttling is retried a limited number of times; the selected image is processed locally and is not sent to a Zhatpat application server. This beta supports images up to 512 × 512 pixels and may use device memory during processing. Other planned tools are not yet enabled; before release, this page will explain whether files are processed in your browser or sent to a service.
Website data
The website may receive standard technical information needed to deliver and secure pages, such as browser type, requested page and server logs. If analytics, advertising or other third-party services are added, this page should be updated to identify them and explain relevant choices.
Third-party services
Some future features may rely on third-party libraries or model services. Their involvement and applicable privacy terms will be disclosed before those features are enabled.
Contact
For privacy questions, use the Contact page. A dedicated support email will be published after the project email domain is configured.