wupage

Project Overview

When you need to translate a foreign-language web page, a PDF handout, or video subtitles in one pass, WuPage offers a way to do it inside the browser: body text, images, PDFs and subtitles all produce translations on the original page without copying anything to another site. It is built on Manifest V3 and installs into Edge, Chrome and Firefox; translation can use online services or run offline with Bergamot. Settings, cache and statistics stay in the current browser, and images, video frames and PDF files are never uploaded to translation services.

Item	Value
Hosts	Microsoft Edge, Google Chrome, Firefox
Extension manifest	Manifest V3
Current version	1.9.0
License	GPL-3.0-only

One extension covers web pages, PDFs, images and video subtitles.

Supported Hosts

• Microsoft Edge: installs from the store or by loading an extracted folder
• Google Chrome: shares the Chromium build with Edge, load wupage-<version>-chrome.zip manually
• Firefox: requires Firefox 154 or later, and the release ZIP is unsigned for testing
• Chromium browsers: can install wupage-<version>-chromium.crx signed with a fixed key, extension ID fgeedjpifhmkdehbgdmngbehdfcflblc

Key Features

• Full page translation: translates body text, dynamically loaded content and embedded frames, with one-click restore
• Paragraph translation: on desktop hold Ctrl and click a paragraph; on touch devices choose "select paragraph", click again to restore
• Bilingual display: replace mode, original-color comparison (default) and two-color comparison, switchable at any time
• Selection lookup: look up a word or translate selected text, and translate text in input fields and replace it in one click
• Video subtitles: click the "translate" button on the player to translate existing subtitles, with SRT / VTT import and bilingual display
• Paused-frame translation: recognizes and translates text in a paused video frame, requiring video subtitles, image translation and the OCR model
• Image translation: on-device OCR shows the translation in place, and the image never leaves the device
• PDF reading and export: opens a PDF and translates around the current page, with editable translations and image or searchable export
• Translation exclusions: keeps chosen areas in the original language by site and CSS selector
• Settings backup: exports a .toml file and imports it after edits

Install and Enable

1. Edge: click "Get" in the extension store, then pin it to the toolbar
2. Chrome or Edge manual install: download wupage-<version>-chrome.zip or wupage-<version>-edge.zip, extract it, open chrome://extensions, enable developer mode and choose "Load unpacked"
3. Firefox: install from Firefox Add-ons, or load the ZIP temporarily in about:debugging
4. Refresh the pages already open, then click the toolbar icon to open the popup
5. Choose a service under "Settings → Translation services", save it, then select it in the popup
6. Install the OCR model for image or frame translation, and download language models for offline translation

Permissions and Scope

• Page access: the <all_urls> host permission reads text to translate, and the original can be restored afterwards
• Background and offscreen: storage keeps settings and cache, while offscreen handles OCR and PDF work
• Shortcut: Alt+Shift+Y triggers full page translation by default and can be changed or disabled in settings
• Data boundary: online services receive only the text and credentials they need; images, video frames and PDF files are not uploaded
• On-device recognition: OCR runs inside the ocr-sandbox.html sandbox using PaddleOCR JS 0.4.2 and onnxruntime-web 1.30.0
• Backup file: .toml exports omit keys by default, and files over 5 MiB or malformed ones are rejected

Notes

• Free services are fragile: preset Microsoft, Google and other services may rate-limit or change their APIs, and failures are not rerouted automatically, so retry or switch manually
• Model size: a local translation model takes about 35–55 MB per language direction, and non-English pairs may go through English
• PDF text layer: scanned PDFs have no text layer and cannot be translated directly
• Font install: searchable PDF export needs the full Noto font package, and the import checks file size and SHA-256
• Subtitle prefetch: subtitles are translated sentence by sentence with about 45 seconds prefetched, and sites that only expose on-screen captions cannot be prefetched