Skip to content
Nappai Documentation

Easy OCR

Easy OCR recognises the text that appears inside an image or a scanned PDF and returns it as plain text. This lets you work with information that only exists “as a picture”: a scanned invoice, a receipt, a screenshot or a sign.

Recognition runs inside NappAI, without sending the image to an external service and without a credential. If the file is a PDF, it reads all its pages one after another.

When to use it

  • You receive scanned invoices or delivery notes and need their text to pull out amounts, dates or order numbers.
  • You want to classify or summarise documents that arrive as images.
  • People send you screenshots of an error or an order and you want to read what they say.
  • You need to pass the text of a scanned PDF to an agent or an extraction component.

Configuration

FieldWhat it does
Input ImageThe image or PDF you want to read. Accepts PNG, JPG, JPEG, BMP, GIF, TIFF and PDF. You usually connect it to the output of a component that loads or receives files.

Advanced options

FieldWhat it does
LanguagesLanguages the text is written in, separated by commas, for example es,en. Default: en,es. Setting the right language improves recognition.
Paragraph ModeGroups nearby lines into paragraphs so the text is not broken into many short lines. On by default.
Min ConfidenceMinimum confidence, from 0 to 1, needed to keep a line. Lines recognised with less confidence are dropped. Default: 0.5. Raise it if meaningless characters appear; lower it if text from blurry documents goes missing. Only applies with Paragraph Mode off: grouped paragraphs carry no confidence, so nothing is dropped.
Detailed OutputAdds the detail of each recognised fragment at the end of the text: its position in the image and how confidently it was read (the confidence only with Paragraph Mode off). Off by default.
Single Paragraph OutputJoins all the text into a single paragraph. If you turn it off, each recognised fragment stays on its own line. On by default.

Output

The Extracted Text output is a message with the recognised text. If no text is found, it returns No text detected in image. You can connect it to an agent, to an extraction component such as Entities extraction, to a summary or to any component that takes text.

Example

A purchasing department receives suppliers’ delivery notes by email as photos.

  1. Connect the email attachment to Input Image.
  2. In Languages, keep en,es, or type only es if all notes are in Spanish.
  3. Connect Extracted Text to an agent instructed to pick out the supplier, the date and the total amount.

Good practices

  • Use sharp, well-lit images: the better the image, the better the result.
  • Turn off Single Paragraph Output when the document has tables or lists and you want to keep the lines.
  • For complex documents (tables, forms, handwriting), also try the components in the OCR section, such as Gemini OCR.