OCR Text Extractor

Read text off images and scanned PDFs with OCR in 19 languages. Every page comes back with the recognised text and an engine confidence score, and every word can come back with its position on the page. No API key, no external OCR service, no per-call quota.

What is OCR Text Extractor?

OCR Text Extractor turns pictures of text into text you can search, index or feed to a model. Point it at PNG, JPG, WEBP, TIFF, BMP, GIF or HEIC files, or at a scanned PDF, and it returns the text page by page.

Two things make it different from a plain OCR wrapper. It tells you how sure it is — an average confidence per page and per word — so you can route doubtful scans to a human instead of trusting them silently. And it returns word boxes: the position of every recognised word, which is what you need to highlight text on the original image or to build a parser that reads fields by location.

What you give it	What you get back
A photo or screenshot of text	Recognised text with a confidence score
A scanned PDF	Text page by page, only the scanned pages billed
A PDF that already has text	The existing text layer, copied for free
A batch of mixed files	One row per file, same fields for all of them

What data can OCR Text Extractor extract?

Field	Type	Example
source	string	https://example.com/invoice.png
format	string	image or pdf
pageCount	integer	12
pagesRead	integer	12
ocrPageCount	integer	9
language	string	eng
text	string	INVOICE 2026-114\nMalina Data…
wordCount	integer	26
charCount	integer	184
confidence	number	93.7
pages	array	[{"number": 1, "text": "…", "confidence": 93.7, "source": "ocr"}]
words	array	[{"text": "INVOICE", "confidence": 95, "left": 124, "top": 168, "width": 174, "height": 32}]
status / errorCode / errorMessage	string	ok, or why a file failed
warnings	array	["Only the first 50 selected page(s) were processed."]
extractedAt	string	2026-09-05T09:14:02+00:00

Each page says whether its text came from OCR or from a text layer the PDF already had, in pages[].source.

How much does OCR cost?

Event	Price	When it is charged
File processed	$0.002	Once per file that produced text
Page read with OCR	$0.009	Per page that actually needed OCR

Pages that already contain text are not charged as OCR pages, and files that fail are returned as error rows at no cost. A 20-page PDF where only 6 pages are scans costs $0.002 + 6 × $0.009 = $0.056, not the price of 20 pages.

Job	Cost
1,000 scanned photos	$11.00
100 scanned PDFs, 10 pages each	$9.20
500 mixed PDFs, a third of pages scanned	$16.00

Comparable OCR Actors charge $15 per 1,000 scanned pages. The Apify Free plan includes $5 of monthly usage, which is about 550 scanned pages.

How to read text from a scan

1. Click Try for free and sign in to Apify.
2. Paste image or PDF links into Image or PDF URLs, or use Upload a file.
3. Pick the OCR language — this matters more than any other setting for accuracy.
4. Switch on Word positions and confidence if you need boxes.
5. Press Start, then read the Output tab or export as JSON, CSV or Excel.

⬇️ Input

{
  "urls": ["https://example.com/scanned-invoice.pdf"],
  "language": "eng",
  "includePages": true,
  "includeWordBoxes": false,
  "useExistingTextLayer": true
}

Picking the language

Accuracy on a good scan is mostly a question of the right language pack. Nineteen are installed: English, German, French, Spanish, Portuguese, Italian, Dutch, Polish, Russian, Ukrainian, Turkish, Indonesian, Vietnamese, Arabic, Hindi, Chinese (simplified and traditional), Japanese and Korean.

If you do not know the language in advance, switch on Detect the script automatically. The page is inspected first and the language is switched when the script is clearly Cyrillic, Arabic, Chinese, Japanese, Korean or Devanagari.

Filtering out unreliable words

minWordConfidence drops words the engine is less sure about than the number you give, from 0 to 100. On a clean scan almost everything sits above 90; on a phone photo of a receipt, setting 60 removes most of the nonsense.

Skipping pages that do not need OCR

useExistingTextLayer is on by default: a PDF page that already contains text is copied rather than re-recognised. That is faster, more accurate and cheaper. Switch it off when a PDF has a bad text layer and you want the picture read instead.

⬆️ Output

{
  "source": "https://…/sample-scan.png",
  "format": "image",
  "pageCount": 1,
  "pagesRead": 1,
  "ocrPageCount": 1,
  "language": "eng",
  "text": "INVOICE 2026-114\nMalina Data\nJalan Raya Uluwatu 88, Jimbaran, Bali\nConsulting services 1,250.00 USD",
  "wordCount": 26,
  "confidence": 93.7,
  "pages": [
    { "number": 1, "text": "INVOICE 2026-114…", "confidence": 93.7, "wordCount": 26, "source": "ocr" }
  ],
  "status": "ok",
  "extractedAt": "2026-09-05T09:14:02+00:00"
}

With Word positions and confidence switched on, each page also carries its words:

{ "text": "INVOICE", "confidence": 95, "left": 124, "top": 168, "width": 174, "height": 32 }

Coordinates are in pixels of the rendered page, measured from the top-left corner.

Error rows

errorCode	Meaning
network-error / http-error	The file could not be downloaded
too-large	The file is above the size limit
unsupported-format	The file is neither an image nor a PDF
corrupt-file	The file is damaged
no-text-found	Nothing readable was found — usually the wrong language or a very low-resolution scan

Use cases

Digitising paperwork

Invoices, permits, contracts and old records that only exist as scans become searchable text, with a confidence score that tells you which ones need a human.

Feeding scans to an LLM

A model cannot read a JPEG of a contract. Run it through here first and pass the text, with the low-confidence words filtered out.

Building a document parser

Word boxes let you read a field by where it sits on the page — the invoice number in the top right, the total above the signature — rather than by guessing from a wall of text.

Screenshots and photos

Text in screenshots, product labels and signs is read the same way as a document scan.

Integrations

• API: POST https://api.apify.com/v2/acts/lergassy~ocr-text-extractor/runs
• Python: ApifyClient(token).actor('lergassy/ocr-text-extractor').call(run_input={'urls': [...]})
• JavaScript: await client.actor('lergassy/ocr-text-extractor').call({ urls: [...] })
• n8n, Make, Zapier, Google Sheets: run the Actor and map the dataset onward.
• MCP: available through the Apify MCP server as a callable tool.
• Webhooks: fire a webhook when a run finishes.

🤖 For AI agents and LLM apps

{ "urls": ["https://example.com/scan.png"], "language": "eng" }

• One row per file; status is ok or error, so a failed file needs no log parsing.
• confidence gives an agent a reason to ask for a better scan instead of acting on noise.
• pages[].source says whether text was recognised or copied from a text layer.
• Files can be passed as base64, so an agent does not need to host anything.
• Failed files are never charged.

❓ FAQ

How accurate is it?

On a clean 200-DPI scan in the right language, high — the sample invoice in this README reads at 94% average confidence. On phone photos, handwriting or heavy stamps, accuracy drops, which is exactly why every result carries a confidence score instead of pretending to be certain.

Does it read handwriting?

Not reliably. The engine is built for printed text.

What is the difference from Document Text Extractor?

Document Text Extractor is for documents that already contain text — PDF, Word, Excel, PowerPoint — and turns them into Markdown, tables and RAG chunks. This Actor is for pictures of text, and returns confidence scores and word positions. Use that one for files, this one for scans.

Do I need an API key for the OCR engine?

No. The recognition runs inside the Actor, so there is no external service, no key and no per-call quota.

Can I use it with the Apify API or an MCP server?

Yes to both, like any Apify Actor.

What is the largest file it can read?

50 MB by default, adjustable to 200 MB. Page count per file is capped at 50 by default so a long scan cannot surprise you.

Your feedback

Missing a language, or a scan that reads badly? Open an issue on the Issues tab — issues are answered quickly.

You might also like

Actor	What it does
Document Text Extractor	PDF, Word, Excel and PowerPoint to Markdown, tables and RAG chunks
Software Reviews Scraper	Capterra, Software Advice, GetApp and TrustRadius reviews in one schema
Email & Phone Verifier	Validate e-mail addresses and phone numbers, find contacts on a website
US Business Leads	New business registrations from US state and city registries