vverity/inpainttranslator-edge-chrome

InpaintTranslator is a browser extension for Chrome and Edge that translates text embedded inside images on web pages. Instead of showing the original foreign-language text, it erases it and redraws the image with the translated words painted in at the same position. The README is written in Chinese. The way it works is: you hover over an image on any web page, click a small blue dot that appears, and the extension sends the image to a local server running on your own computer. That server uses PaddleOCR (a text-recognition tool) to find and read the text in the image, then passes it to Ollama (a tool for running AI language models locally) to produce the translation. The server paints the translated text back into the image at the original positions and returns the result, which the extension displays in place of the original image. Clicking again toggles back to the original. A keyboard shortcut (Alt+A) is also available to trigger the translation without using the mouse. Setting it up requires two steps. First you run the Python server locally by installing the listed packages and starting ocr_server.py, which listens on port 8000. Second you load the extension into the browser in developer mode by pointing it at the project folder. The extension settings let you change the server address and target language. The project depends on Python with paddleocr, fastapi, opencv-python, and pillow, plus Ollama installed separately with a model already downloaded. It also mentions support for the reading direction used in Japanese and Chinese comics (right-to-left, top-to-bottom). The repository is a personal-use project with an MIT license.