Skip to content

OCR Providers

Torto’s PDF OCR / Reflow feature turns a scanned PDF into reflowable, searchable text. It relies on a cloud OCR service. Two providers are currently supported:

Provider Highlights What you enter in Torto
PaddleOCR (default) Official Baidu PaddlePaddle service with a free daily quota on AI Studio Access Token
MinerU Document parsing by OpenDataLab, great on complex layouts API Token

Open Settings → OCR in Torto to pick a provider and enter credentials. Tokens are stored in your OS credential manager, never in config files.

PaddleOCR’s cloud service is hosted on PaddlePaddle AI Studio. Every registered user gets a personal API endpoint and access token, with a free daily quota (see the official page for current limits).

  1. Open the PaddleOCR task page on AI Studio and sign in with a Baidu account (register one first if needed).
  2. Click the 「API」 button in the top-left corner of the page.
  3. In the API usage example, find PaddleOCR-VL and copy the TOKEN — a 40-character hexadecimal access token. (The page also shows a personal API_URL; Torto has the service address built in, so you don’t need it.)
  1. Go to Settings → OCR and select PaddleOCR.
  2. Paste the TOKEN into “Access Token”.
  3. Pick a model — PaddleOCR-VL-1.6 is the default; PaddleOCR-VL-1.5 and PaddleOCR-VL are also available.
  4. Open any scanned PDF and run OCR. Results are cached locally, and you can switch between the Original and Reflow views.

MinerU is a document-parsing service by OpenDataLab that handles multi-column layouts, tables and formulas well.

  1. Open mineru.net, sign up and log in.
  2. Go to User Center → API Management (or directly to the API management page) and create an API Token.
  3. Copy the token and keep it safe.
  1. Go to Settings → OCR and select MinerU.
  2. Paste the token into “API Token” (the API address is built in — nothing else to fill in).
  3. Pick a model: vlm (default, better quality) or pipeline (faster).

OCR jobs stuck or failing? Check that the token is correct and unexpired. For PaddleOCR, confirm you haven’t used up the daily free quota. Your network must be able to reach the provider’s API domain.

Where are results stored? OCR output is cached locally per page — the same book never consumes quota twice, and switching between Original/Reflow views makes no new requests.