OCR Image to Text — extract text from images and scanned PDFs

For developers, analysts and AI/RAG pipelines that need the text inside images. Give it public image URLs (PNG, JPEG, WebP, TIFF, GIF) and optionally scanned PDF URLs, and get one row per image with the recognized text, language, confidence, word count and image size. Runs the open-source Tesseract OCR engine inside the Actor: no external OCR API, no API key, nothing to upload. $0.004 per image read. Leave the input empty for a free demo.

Use cases

• Turn screenshots, photos of signs, receipts, invoices and slides into searchable text
• Read scanned PDFs (each page is rasterized and read as an image) before indexing them for search or RAG
• Add alt-text or keyword data to image libraries and product catalogs
• Get line-level bounding boxes to locate text in an image (lines layout)
• Pre-process a batch of image links from a scraper before sending the text to an LLM

How it works

• Each URL is downloaded once. Images are converted to grayscale with contrast stretching and small images are upscaled (both can be turned off), then read with Tesseract.
• Languages: English, German, French, Spanish, Portuguese, Italian, Japanese and Simplified Chinese. Combine up to 4 (for example English + German). Fewer languages are faster and more accurate.
• Scanned PDFs: pages are rendered at 200 dpi and read one by one, up to maxPdfPages (default 5, max 50). Each page is one row with its page number and counts as one image for pricing.
• Output layout: text (plain text, paragraphs separated by a blank line) or lines (the same plus each line's text, bounding box and confidence).
• Pay only for results: an image or PDF page is charged only when at least one word was recognized. Failed downloads, unreadable files, password-protected PDFs and images without text are free. The free demo (empty input, two built-in sample images) is not charged.
• Safe by design: URLs that resolve to internal addresses (localhost, private ranges, link-local such as 169.254.169.254) are never contacted, including redirect targets (blocked_private_address). Only http and https, max 5 redirects, 30 s per download, size cap (default 10 MB, max 25 MB), images above 40 megapixels are skipped.

Input

{
  "urls": ["https://example.com/receipt.jpg", "https://example.com/scan.pdf"],
  "languages": ["eng", "deu"],
  "outputLayout": "text",
  "maxImageSizeMB": 10,
  "maxPdfPages": 5,
  "grayscale": true,
  "upscaleSmall": true
}

Output

{
  "url": "https://example.com/receipt.jpg",
  "page": null,
  "text": "Invoice No. 2026-0042\nTotal due: USD 1,234.50",
  "language": "eng+deu",
  "confidence": 93.4,
  "wordCount": 8,
  "width": 1000,
  "height": 220,
  "error": null
}

confidence is the mean word confidence from 0 to 100. page is null for images and the 1-based page number for PDFs. With "outputLayout": "lines" each row also has lines: [{"text": "Invoice No. 2026-0042", "left": 40, "top": 40, "width": 520, "height": 44, "confidence": 95.1}] (pixels of the original image). error is e.g. http_404, not_an_image, file_too_large, timeout, too_many_redirects, unreadable_image, image_dimensions_too_large, unreadable_pdf, encrypted, ocr_timeout, blocked_private_address, invalid_url. A failed file produces one row with empty text. A SUMMARY record in the key-value store has the counts (rows, ok, errors, imagesCharged, words).

Pricing

$0.004 per image read (pay per event, image). A scanned PDF costs $0.004 per page read. Example: 100 images cost $0.40, a 5-page scanned PDF $0.02. Images without any recognized text and failed files are free. For comparison, other OCR Actors in the Apify Store charge roughly $0.003 to $0.005 per image or page.

Limits and notes

• Printed text works best. Clean, horizontal, high-contrast text (documents, screenshots, signs, receipts) is read well. Handwriting is not supported, and stylized fonts, curved text, very low resolution and heavy noise give poor results. Check confidence.
• Tables and multi-column layouts are returned as lines in reading order, not as a structured table.
• Japanese and Chinese text is slower than Latin scripts. Pick only the languages you need.
• Only the files you list are downloaded; links inside them are not followed. Files behind a login return http_401 or http_403.
• Only process images you are allowed to download. The text belongs to the rights holder of the image.

More tools from the same developer

• PDF Extractor — PDF to text, tables and metadata from public PDF URLs, page by page
• RAG Chunker — split scraped pages into chunks for RAG and vector stores
• Website Screenshot API — screenshots and PDFs of web pages: full page, mobile, element, no cookie banners
• Google Sheets Export — export an Apify dataset to Google Sheets
• Bulk URL Status Checker — bulk HTTP status and redirect-chain checker for URL lists