Image to Text OCR — photos, scans & screenshots to text

Send image URLs and get the text back: plain text plus per-line bounding boxes and confidence scores, detected script, automatic rotation fixing. Tesseract runs inside the Actor — no API key, no external OCR service, no per-provider markup, and no scraping of anyone's OCR endpoint that can break under you.

What you get per image

• text — plain text, one OCR line per output line, in reading order (Arabic and other RTL scripts come out in correct logical order).
• lines — every line with its pixel bounding box and mean confidence: {text, x0, top, x1, bottom, conf}.
• words — optional per-word detail (include_words), off by default.
• script (latin / arabic / mixed), language_used, mean_confidence, dimensions, applied rotation.

Languages

English, Arabic (plus mixed Arabic+English), French, German, Spanish, Portuguese, Italian, Dutch, Turkish, Russian, Persian, Urdu. auto detects the script per image and picks the model; setting the language explicitly is always more accurate when you know it.

Pricing — you only pay for delivered text

$0.0015 per image that delivers at least one recognized word.

An image that fails to download, cannot be decoded, or contains no recognizable text is never billed. No start fee, no per-run fee.

Sources and limits

• Direct image file URLs: jpg, png, webp, tiff, bmp, gif.
• Up to 50 URLs per run, 25 MB per image.
• Very large images are automatically downscaled to ~40 MP before OCR (accuracy above ~300 dpi equivalent is flat; this bounds cost and time).
• Animated GIFs and multi-page TIFFs use their first frame.
• Phone photos stored rotated (EXIF) are handled; sideways scans are detected and fixed when auto_rotate is on.
• This actor does not crawl web pages for images — send direct file URLs you have the right to process.

Tips

• Screenshots and UI captures: already sharp — leave defaults.
• Receipts/documents as photos: keep auto_rotate on.
• Noisy photos: set min_word_confidence to 60+ to drop guesses.
• Need scanned PDFs instead of images? Use amanatools/pdf-ocr-extractor — same engine, built for PDFs, with layout preservation.