rapidai/rapidocr — reverse-engineered prompt

Build me a fast, fully offline OCR project that can read text from images with very little setup. I want it to feel open source and practical, with Chinese and English working out of the box, and a clear path for adding more languages from available models. The main experience should be a simple Python entry point where I can pass an image file or URL and get back recognized text plus bounding boxes, and save a visualized result image too.

Please make it work across common desktop systems, keep it lightweight, and structure it so the OCR engine can be used from more than just Python later. If this repo already includes a small web demo or API, wire those up so I can quickly try it in a browser as well. Add sane docs, quick start steps, and one or two sample images so someone can test it immediately. If anything is unclear, look up the current project docs online and make the implementation match the intended RapidOCR experience.