Image to Text OCR – extract text from images in bulk (100+ languages)

Image to Text OCR reads text from any number of images and returns it as clean, searchable data. Paste image links – PNG, JPG/JPEG, WebP, GIF, TIFF, BMP, AVIF or SVG – and get the plain text, a confidence score, and optionally every line and word with its bounding box. It runs the open-source Tesseract OCR engine inside the Actor, so there is no API key, no LLM, no per-page cloud OCR bill, and your images are not sent to any third-party AI service.

Images are auto-rotated, contrast-normalised and upscaled before recognition, which noticeably improves accuracy on phone photos, screenshots and small images.

What can this OCR tool do?

• 🖼️ Bulk image OCR – hundreds or thousands of image URLs per run; download of the next image overlaps with OCR of the current one
• 🌍 100+ languages – English built in; deu, fra, spa, ita, por, nld, pol, rus, ukr, ara, hin, jpn, kor, chi_sim, chi_tra… and combinations like eng+deu
• 🧭 Layout modes – automatic (documents, screenshots), single block, single column, sparse text (signs, memes, UI, product photos), single line (labels, captions)
• 📐 Lines & words with coordinates in original-image pixels – for redaction, form parsing, highlighting or training data
• 🎯 Confidence filter – drop low-confidence words to remove noise from photos
• 🛠️ Robust – EXIF rotation, transparent PNGs, CMYK JPEGs, animated GIFs/multi-page TIFFs (first frame), 25 MB size cap, retries with back-off
• 💸 You only pay for images where text was found – broken links, non-images and blank images are free

Who uses it?

• Ops & finance teams – receipts, invoices, delivery notes and screenshots → searchable text
• E-commerce & marketing – read text on product images, banners and ads (with permission) for QA and cataloguing
• Developers & AI builders – OCR step for RAG pipelines, agents and n8n/Make/Zapier automations
• Researchers & archivists – digitise scanned letters, posters and historical documents
• Accessibility – generate text for images that lack alt text

How much does image OCR cost?

$4 per 1,000 images with text ($0.004 per image; $0.0035 on Silver, $0.003 on Gold+) plus $0.001 per run. No charge for images that fail to download, aren't images, exceed the size limit, or contain no text.

Job	Price
100 screenshots	~$0.40
1,000 receipts	~$4
10,000 product images	~$40

Use Max images or Max cost per run to cap spending. Apify Free plan users get $5 of monthly credit – over 1,000 images.

Working with PDFs (including scanned PDFs)? Use our PDF to Markdown & Text Converter – it OCRs scanned pages automatically.

How to use it

1. Click Try for free.
2. Add image URLs (or upload files to an Apify key-value store and paste their links).
3. Choose the language (e.g. eng, deu, eng+fra) and, for photos of signs/UI, the Sparse text layout.
4. Click Start. Export as JSON, CSV, Excel or via API.

Input example

{
  "imageUrls": [
    { "url": "https://example.com/receipts/2026-10-01.jpg" },
    { "url": "https://example.com/screens/dashboard.png" }
  ],
  "language": "eng",
  "pageSegmentationMode": "auto",
  "includeLines": true,
  "includeWords": false,
  "minWordConfidence": 0
}

Output example

Real output from a 1961 typewritten letter (Wikimedia Commons) – note OCR is not perfect on old typewriter text ("Anns" for "Arms").

{
  "url": "https://upload.wikimedia.org/…/Typewritten_letter_from_the_Savage_Arms_Parts_Department_(1961).jpg",
  "text": "Savage Anns Corporation\n\nMOF\nJanuary 26, 1961\n\nCLYDE WALSER\n\nBox 145\nFRANKLINTON, NORTH CAROLINA\n\nDear MR. WALSER: …",
  "confidence": 83,
  "hasText": true,
  "wordCount": 78,
  "lineCount": 21,
  "lines": [{ "text": "CLYDE WALSER", "confidence": 91, "bbox": { "x0": 297, "y0": 499, "x1": 413, "y1": 511 } }],
  "language": "eng",
  "width": 1920,
  "height": 1440,
  "format": "jpeg",
  "sizeBytes": 817937,
  "ocrMs": 2260,
  "charged": true
}

Failed items have an error field (e.g. HTTP 404, Not a supported image, Image is larger than 25 MB) and "charged": false. A RUN_SUMMARY record shows totals.

Tips for best accuracy

• Text should be at least ~15 px high – the Actor upscales small images automatically.
• Use Sparse text for photos where text is scattered (signs, packaging, memes, app screenshots).
• Set the right language – accents and non-Latin scripts need their language model.
• Use minWordConfidence: 40–60 to drop garbage from noisy photos.
• Handwriting is not supported well by Tesseract; printed and typed text works best.

FAQ

Is my data sent to an AI provider? No. OCR runs locally inside the Actor with Tesseract (Apache-2.0). Only your image URLs are downloaded.

Is it legal? OCR of images you own or have permission to process is fine. Do not use it to extract personal data you have no lawful basis to process, and respect the terms of the sites hosting the images.

Can I call it from code or AI agents? Yes – Apify API, JavaScript/Python clients, webhooks, Zapier, Make, n8n, or as an MCP tool.

Questions or a language that doesn't work? Open an issue on the Issues tab.

Need the data done for you?

If you'd rather not set up runs yourself, the UK company behind these tools (GPT Shop Ltd) offers a fixed-price done-for-you scrape. Tell us the website and the fields you need, and you get a clean CSV or Google Sheet within 48 hours of the confirmed brief, up to 5,000 rows, for £149 or $179. If we can't get the data, you're refunded in full. See the done-for-you scraper service