Image to Text OCR: images & scanned PDFs, pay per page

Turn images and scanned PDFs into clean text by URL, up to 50 files per run. $0.005 per OCR'd page, no start fee. Weak on handwriting. Built for automation pipelines (digitizing paper documents, RAG ingestion of scans, form and receipt data entry, search indexing).

Use cases

• Digitizing paper documents and scanned PDFs into searchable text
• RAG ingestion of scanned documents
• Form, label and receipt data entry
• Extracting text from images in a pipeline (png, jpg, tiff, webp, bmp, gif)

What you get

For each file, one dataset item:

• status: ok, download_failed, unsupported_format, invalid_file, or no_text_found
• fileType: detected from the file's content (magic bytes), never the URL extension: pdf, png, jpg, tiff, webp, bmp, or gif
• pages[]: per-page text, meanConfidence (mean word confidence, 0-100) and wordCount
• fullText, totalChars, pageCount, pagesProcessed, truncated
• processingMs, and error with a plain-language reason when something fails

PDF pages are rendered with pypdfium2 at your chosen DPI (default 200) and OCR'd with Tesseract. Images (png, jpg, tiff, webp, bmp, and the first frame of gif) are OCR'd directly.

Pricing: you only pay for pages that worked

One page-ocred event per processed page of a file that came back ok ($0.005). Failed downloads, unsupported or invalid files, and OCR runs that found zero words (no_text_found) are reported in the output but never charged. There is no start fee. Files with more pages than your limit are truncated and charged only for the pages actually processed. The run checks your spending limit before each file, processes at most the pages your remaining limit covers, and stops cleanly when it is reached.

Accuracy: Tesseract OCR, open source

The engine is Tesseract, the standard open-source OCR. It is good at printed text: documents, scans, screenshots, forms, labels and photos of text taken straight-on at reasonable resolution, typically at 200-300 DPI. It is weak on handwriting, stylized or tiny fonts, and low-resolution or blurry photos — expect lower confidence and occasional misreads there. The per-page meanConfidence field tells you how sure the engine was, so you can route low-confidence pages to review. Six languages are available out of the box: English, Spanish, French, German, Portuguese and Italian.

Limits (by design)

• Max 50 URLs per run, max 25 MB per file, max 200 pages per file (default 20).
• Supported inputs: png, jpg, tiff, webp, bmp, gif (first frame), and PDF. Anything else is reported as unsupported_format and not charged.
• Only public http(s) links; local and private-network addresses are refused.
• No table structure or layout detection: output is plain text in reading order.

FAQ

Does it work on scanned PDFs? Yes — PDF pages are rendered at your chosen DPI (default 200) and OCR'd with Tesseract.

Is it good on handwriting? It is weak on handwriting, stylized or tiny fonts, and low-resolution or blurry photos — expect lower confidence and occasional misreads there. The per-page meanConfidence field tells you how sure the engine was.

Which languages are supported? Six out of the box: English, Spanish, French, German, Portuguese and Italian.

Do I pay for pages that find no text? No. Failed downloads, unsupported or invalid files, and no_text_found results are reported but never charged; you pay $0.005 only for pages actually processed in a file that came back ok, with no start fee.

How are multi-page files charged? Per processed page, up to your limit; truncated files are charged only for the pages actually processed.

What are the limits? Up to 50 public http(s) URLs per run, 25 MB per file, 200 pages per file (default 20).

Input example

{
  "urls": ["https://example.com/scan.png"],
  "language": "eng",
  "maxPagesPerFile": 20,
  "dpi": 200
}