OCR Text Extractor - Images & Scanned PDFs to Text

Turn images and scanned PDFs into clean, structured text. Give the actor a list of links to PNG, JPG, TIFF, WEBP, BMP or GIF images or PDF files and get back the recognised text, per-page text, confidence scores, optional word bounding boxes, and a status for every file - as JSON, CSV or Excel.

Powered by Tesseract OCR 5 (LSTM engine) with 28 languages built in, plus automatic text-layer detection: PDFs that already contain selectable text are read exactly and instantly, and only true scans are rasterised and OCR-ed. No API key, no OCR subscription, no browser - it all runs inside the actor.

Why this actor exists: the OCR / image-to-text niche on Apify draws hundreds of monthly users across tools that either call an external OCR API (and pass on its cost) or only read PDFs that already have a text layer. This one does real OCR on scans and photos, reads born-digital PDFs for free, and prices per page at a fraction of a cent.

What you get per file

Field	Description
text	The full recognised text of the file (all pages joined)
pages[]	Each page separately: page, text, wordCount, meanConfidence, method (ocr or text-layer)
words[]	Optional: every word with confidence (0-100) and its box left, top, width, height in pixels
meanConfidence minPageConfidence	OCR confidence (0-100) averaged over words, and the weakest page - use it to route low-confidence scans to review
pageCount pagesProcessed method	Pages in the file, pages actually processed, and whether OCR, the PDF text layer or a mix was used
wordCount charCount lineCount width height sizeBytes mimeType	Size and length metrics; image dimensions; the detected file type
languages pageSegmentation durationMs	The recognition settings used and the processing time
status error	ok, or failed with the reason (bad link, not an image/PDF, too large, unreadable) - failures are never charged

Input

{
  "imageUrls": [
    "https://tesseract-ocr.github.io/tessdoc/images/eurotext.png",
    "https://www.irs.gov/pub/irs-prior/f1040--1913.pdf"
  ],
  "languages": ["eng"],
  "pageSegmentation": "auto",
  "preferTextLayer": true,
  "includePageText": true,
  "includeWords": false
}

• Image / PDF URLs - direct links to image or PDF files, one per line. Leave empty for a small free sample.
• Import URLs from a file - paste a whole list, or link a public .txt/.csv, a Google Sheet/Drive share link, or an Apify key-value-store record. Blank lines, # comments, quotes and extra CSV columns all parse; a file that cannot be read says so and charges nothing.
• Languages - pick one or more of 28 languages (English, German, French, Spanish, Italian, Portuguese, Dutch, Polish, Russian, Ukrainian, Turkish, Arabic, Hebrew, Hindi, Japanese, Korean, Chinese simplified and traditional, Czech, Swedish, Danish, Norwegian, Finnish, Hungarian, Romanian, Vietnamese, Indonesian, Thai). Fewer languages = faster and more accurate.
• Page layout - auto for ordinary documents; sparse for screenshots, receipts and labels; single_line for captchas, serial numbers and labels; single_block for one clean paragraph; columns for single-column text with mixed sizes.
• Use the PDF text layer when present - born-digital PDFs are read exactly instead of OCR-ed (default on). Each page reports its method.
• PDF render resolution - 200 DPI default; 300 for small print; 150 for speed.
• Include per-page text / Include word boxes - shape the record for your use case (search index, redaction, form parsing, layout reconstruction).
• Max pages per PDF - default 20; set 0 for all pages. Max file size - default 30 MB.
• Output fields - tick only the fields you want for a lean CSV/Sheets export.

Example output

{
  "type": "ocr",
  "url": "https://tesseract-ocr.github.io/tessdoc/images/eurotext.png",
  "filename": "eurotext.png",
  "sourceType": "image",
  "status": "ok",
  "error": null,
  "sizeBytes": 31481,
  "mimeType": "image/png",
  "pageCount": 1,
  "pagesProcessed": 1,
  "method": "ocr",
  "languages": ["eng"],
  "pageSegmentation": "auto",
  "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.\n...",
  "wordCount": 68,
  "charCount": 421,
  "lineCount": 8,
  "meanConfidence": 91.4,
  "minPageConfidence": 91.4,
  "width": 1024,
  "height": 480,
  "pages": [{ "page": 1, "method": "ocr", "text": "...", "wordCount": 68, "charCount": 421, "lineCount": 8, "meanConfidence": 91.4 }],
  "durationMs": 1830,
  "scrapedAt": "2026-09-02T06:00:00.000Z"
}

Use cases

• Digitise scanned archives - contracts, invoices, receipts, forms, letters, historical records - into searchable text.
• Invoice / receipt / form data capture - turn photos and scans into text you can parse with rules or an LLM; use words[] boxes to locate fields.
• Screenshot and image text extraction - pull text from screenshots, memes, slides, charts, product labels and packaging.
• Make PDFs searchable - extract text from scanned PDFs for search indexes, RAG pipelines and knowledge bases; born-digital PDFs are read for free via the text layer.
• Compliance and review - route pages under a confidence threshold to a human reviewer using meanConfidence / minPageConfidence.
• Multilingual document processing - 28 languages, combinable for mixed-language documents.

Tips for the best accuracy

• Scan at 200-300 DPI; photos should be sharp, well-lit and not skewed.
• Choose the exact language(s); adding languages the document does not contain lowers accuracy.
• Use sparse layout for receipts, screenshots and labels; single_line for one-line labels and codes.
• Check meanConfidence: below ~70 usually means a poor scan, the wrong language, or the wrong layout mode.

Use with AI assistants (MCP)

This actor is available through the Apify MCP server, so assistants like Claude can call it as a tool - "extract the text from these three scanned invoices" or "OCR this screenshot in German" - and get structured text back.

Agent-ready: autonomous payments (x402 & Skyfire)

This actor is agent-ready - AI agents can discover it, run it, and pay for it autonomously, with no Apify account and no human in the loop. It uses pay-per-event pricing and limited permissions, so it qualifies for Apify's agentic-payment standards:

• x402 - an open, HTTP-native payment protocol. Agents pay per run in USDC on the Base network directly through the Apify MCP server - no account, no API key.
• Skyfire - agent-to-service payments for fully autonomous AI-agent workflows.

Building an AI agent, MCP tool, or autonomous data pipeline? This tool is ready to plug in and pay as it goes.

Pricing

Pay-per-event, tiered by volume: $0.003 per page on the free tier, dropping to $0.00075 per page at the top tier. An image is one page; a PDF is charged per page actually processed (text-layer pages included). Failed downloads, unsupported files and unreadable pages are never charged.

FAQ

Does it work on photos of documents, not just scans? Yes - sharp, well-lit photos work well. Skew and blur reduce confidence.

What about handwriting? Tesseract is built for printed text. Neat handwriting sometimes works; cursive generally does not.

How are PDFs handled? Every page is checked for a text layer first. Pages with real text are read exactly (method: "text-layer"); scanned pages are rasterised at your DPI and OCR-ed (method: "ocr"). A file with both reports method: "mixed".

Which formats are supported? PNG, JPEG, TIFF (including multi-page), WEBP, BMP, GIF and PDF. The type is detected from the file itself, not the extension.

Can I get the position of each word? Turn on Include word boxes to get words[] with pixel coordinates and per-word confidence.

Is there a size limit? Files over the Max file size (default 30 MB) are skipped with a clear error and no charge. Very long PDFs are capped by Max pages per PDF (default 20, 0 = all).

Does it need a proxy? No. Files are downloaded directly; enable a proxy only if a host blocks the download.

Every run finishes with a clear status message; runs with nothing to process never crash and never bill.