Bulk OCR – Image & Scanned PDF to Text (Tesseract, 100+ Languages)

Extract text from images and scanned PDFs in bulk. Paste a list of image or PDF URLs and get back clean text for every image and every PDF page, with a confidence score and, if you want, every line and word with its bounding box. Powered by the open-source Tesseract OCR engine, so there are no API keys, no OpenAI/Google Vision bills and no per-call quotas.

$1 per 1,000 OCR'd pages. You only pay for pages where text was recognised. Failed downloads, unsupported files and blank pages are listed with a status and cost nothing.

What it does

• 🖼️ Images: JPG, PNG, TIFF (including multi-page TIFF), WebP, BMP, GIF
• 📄 PDFs: scanned or digital, every page is rendered to an image (PyMuPDF) and OCR'd; limit pages per file with maxPagesPerPdf
• 🌍 100+ languages: English, German, French, Spanish, Italian, Dutch, Polish and Portuguese are built in. Any other Tesseract language (Japanese jpn, Chinese chi_sim/chi_tra, Arabic ara, Russian rus, Ukrainian ukr, Hindi hin, Korean kor, Turkish tur, ...) is downloaded automatically on first use. Combine them with +, e.g. eng+fra
• 📐 Three output levels: plain text, text + lines, or text + lines + words, each with x, y, w, h bounding boxes and confidence
• 🧹 Auto preprocessing: grayscale, contrast stretch, upscaling of small images, automatic deskew of slightly crooked scans and binarization. Optional auto-rotate for sideways or upside-down pages
• ⚡ Fast and cheap: a typical page takes 1–4 seconds; runs in 1 GB of memory

Who it's for: anyone digitising scanned documents, invoices, letters, book pages, archive material or screenshots; developers who need OCR inside a pipeline (n8n, Make, Zapier, LangChain, AI agents) without managing Tesseract themselves; RAG builders who need text from scanned PDFs.

Why this actor: most OCR actors in the Store charge $1.50–$12 per 1,000 pages. This one is $1 per 1,000 pages, free for failures and blank pages, supports 100+ languages and returns word-level boxes for layout work.

Honest note on quality

Tesseract is excellent on printed text: scans, PDFs, screenshots, book pages, typed letters (typically 90–99 % word confidence on clean scans). It is mediocre on photos with perspective, glare or busy backgrounds, and poor on handwriting. Tables are read line by line, not as structured tables. Use meanConfidence to spot pages that need a human look. If you need handwriting recognition, a vision LLM or a cloud OCR API is the better (and more expensive) choice.

Input

Field	Default	Description
urls	2 Tesseract test images	Public image or PDF URLs. Type is detected from the file content
languages	eng+deu	Tesseract language codes joined with +. Fewer languages = faster, more accurate
outputLevel	text	text, lines or words (adds bounding boxes)
preprocess	true	Grayscale, contrast, upscale, deskew, binarize
autoRotate	false	Detect and fix 90/180/270° rotated pages
psm	3	Tesseract page segmentation mode: 3 auto, 4 single column, 6 single block, 7 single line, 11 sparse text (screenshots, signs)
maxPagesPerPdf	20	Only the first N pages of each PDF / multi-page TIFF
pdfDpi	200	Render resolution for PDF pages (300 for small print)
minConfidence	0	Drop words below this confidence (0–100)
maxFileSizeMb	50	Skip larger files (free)
maxConcurrency	2	Pages OCR'd in parallel

{
  "urls": [
    "https://raw.githubusercontent.com/mozilla/pdf.js/master/web/compressed.tracemonkey-pldi-09.pdf",
    "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif"
  ],
  "languages": "eng",
  "outputLevel": "words",
  "maxPagesPerPdf": 2
}

Output

One dataset item per image or PDF page. Real example from a test run (default input):

{
  "sourceUrl": "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/eurotext.tif",
  "page": 1,
  "pageCount": 1,
  "sourceType": "image",
  "language": "eng+deu",
  "status": "ok",
  "text": "The (quick) [brown] {fox} jumps!\nOver the $43,456.78 <lazy> #90 dog\n& duck/goose, as 12.5% of E-mail\nfrom aspammer@website.com is spam.\nDer „schnelle” braune Fuchs springt\nüber den faulen Hund. Le renard brun\n«rapide» saute par-dessus le chien\nparesseux. ...",
  "meanConfidence": 92.4,
  "wordCount": 66,
  "lineCount": 12,
  "charCount": 347,
  "width": 1024,
  "height": 800,
  "rotated": 0,
  "deskewAngle": 0.8,
  "processingMs": 1234,
  "error": null
}

With outputLevel: "words" each item also has lines and words arrays. Coordinates are pixels of the original image (for PDFs: of the page rendered at pdfDpi):

"words": [
  { "text": "Trace-based", "confidence": 90, "x": 225, "y": 220, "w": 257, "h": 35, "block": 1, "paragraph": 1, "line": 1 }
]

Field	Meaning
status	ok (text found, charged), empty (no text found, free), error (download/file/OCR problem, free)
page	1-based page number (null when the whole file failed)
pageCount	Total pages in the source file
meanConfidence	Average word confidence 0–100
text	Recognised text; line breaks kept, blank line between paragraphs
deskewAngle, rotated	Corrections applied by preprocessing
error	Reason for error rows, e.g. HTTP 404 when downloading, PDF is password protected

Pricing

Pay per event: $0.001 per OCR'd page (an image counts as one page). 1,000 scanned pages cost $1. Errors and pages without any text are free. Set a maximum cost per run in the run options and the actor stops cleanly when it is reached.

Use via API, n8n, Make or AI agents

API (synchronous, returns the dataset items):

curl -X POST "https://api.apify.com/v2/acts/fast_data_asap~bulk-ocr-image-to-text/run-sync-get-dataset-items?token=YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"urls":["https://example.com/scan.pdf"],"languages":"eng","maxPagesPerPdf":5}'

Python:

from apify_client import ApifyClient
client = ApifyClient("YOUR_TOKEN")
run = client.actor("fast_data_asap/bulk-ocr-image-to-text").call(run_input={"urls": ["https://example.com/invoice.png"], "languages": "eng"})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["page"], item["meanConfidence"], item["text"][:200])

n8n / Make / Zapier: use the Apify node/app, action "Run actor", actor fast_data_asap/bulk-ocr-image-to-text, then read the dataset items.

MCP (Claude, Cursor, other AI agents): add https://mcp.apify.com?tools=fast_data_asap/bulk-ocr-image-to-text as an MCP server and let your agent OCR images and PDFs on demand.

Adding more languages

Just put the code in languages; missing language data is downloaded from the official tesseract-ocr/tessdata_fast repository at the start of the run (a few seconds). The full list of codes is in the Tesseract documentation. If you fork the actor and want a language baked into the image, add its Debian package (e.g. tesseract-ocr-jpn) to the apt-get install line in the Dockerfile.

FAQ

Can it read files from Google Drive or Dropbox? Yes, if you use a direct download link that returns the file itself (not an HTML preview page).

Does it handle digital (non-scanned) PDFs? Yes, they are rendered and OCR'd like scans. For digital PDFs a text extractor is faster and exact, but OCR works when the text layer is broken or missing.

Is my data stored? Files are only downloaded into the run's container for processing; results go to your own Apify dataset.