Readerbox — Kindle Cloud Reader to searchable PDF (OCR)

Readerbox is a desktop Chrome extension that helps you export and save as PDF books you bought from the supported US or Japan store and can open in Kindle Cloud Reader (Kindle for Web). It captures the pages shown on screen, converts them into page images with text, then runs OCR on your computer. Key features - Page images and an invisible searchable text layer stay together in one file, so you can search, select, and copy text. - Japanese vertical text is supported. The recognized text is not a TXT export; it is embedded as the text layer. - Use the result with ChatGPT or NotebookLM, or move it to iPad apps such as Goodnotes to annotate. Large books may need splitting for some services. - Readerbox does not remove DRM, decrypt book files, or obtain ebook files. It works only with pages displayed in your browser. - Readerbox does not upload page images, recognized text, or generated files. For billing, it sends only data needed to verify status, such as install ID, selected plan, book credits, and a per-install book hash. Card details are handled by Stripe. How to use Desktop Chrome 118 or newer on Windows or Mac is required. Open the target book in the supported reader site, click the Readerbox icon, choose a quality setting, and start. Readerbox turns pages, captures them, runs recognition, and builds the file automatically. Keep the reader window in front during capture. If a run stops, captured pages remain available so you can continue. Accuracy, time, and size For Japanese vertical text, standard and high detail were measured at about 1 character error per 100 characters: CER 0.80%. Light was about 6 character errors per 100 characters, CER 6.30%, and it does not read horizontal Japanese. For English books, the standard setting was measured at about 0.5 character errors per 100 characters: CER 0.53%. High detail is not a more accurate OCR setting than standard; it keeps diagrams and small text clearer in the page image. At standard quality, a 200-page Japanese book produces a PDF of about 30MB; the measured case was 195 pages and 33,310,849 bytes. A standard 200-page run takes about 25 minutes, and about 2/3 of that time is text recognition. Services with file-size or page-count limits may require large books to be split. Before capture starts, Readerbox has to return the book to the beginning. Kindle Cloud Reader does not provide a direct jump to the first page, so Readerbox steps back page by page. Measured rewind speed was 0.9-5.9 seconds per page; a 200-page book opened at the end can spend 3-20 minutes just getting back to the start. Opening the book near the beginning first shortens that time. The extension download is about 90MB, including about 85MB of recognition model files. The size is the tradeoff for doing OCR on your own computer. Scope and limits The supported purchase stores and reader domains are Amazon.com / Amazon.co.jp and read.amazon.com / read.amazon.co.jp. Other stores and reader domains are not supported. Books that cannot open in the browser cannot be captured. Please do not use it for books you are reading through Kindle Unlimited or Prime Reading. iPad, phones, and Safari are not supported. You can move the saved PDF to iPad apps for annotation or handwriting. Books that are mostly figures or illustrations may fail to start because page turns cannot be detected. Pricing Your first 3 books are free. After that, making a 4th new book requires a monthly license ($2.99 / month), yearly license ($24.99 / year), or book credits: $1.29 for 1 book, $5.49 for 5 books, or $17.99 for 20 books. Rebuilding the same book is free, and failed books do not count. The seller is Blue Inc. Readerbox is not affiliated with, endorsed by, or sponsored by Amazon or Kindle. The names above are used only to describe the supported service and example uses.