Image to Text OCR: images and scanned PDF OCR, per line

Text from images by URL, up to 200 per run: every line with its confidence and position, the full text joined, image size and format, for screenshots, scans, receipts, menus and product labels, without a browser or an API key. Pay per image with text.

Why this scraper

• Only delivered results are charged. Inputs that fail come back as clear error records at no cost.
• Open-source OCR (PaddleOCR models on onnxruntime) inside the actor, so no external API, no key and no per-call vendor fee: a 640 by 480 screenshot reads in about two seconds. Measured on the platform: eight lines at 0.98 confidence for seven hundredths of a cent.
• Every line separately with its confidence and its box (x, y, width, height), plus the full text joined in reading order, the image size and format. Lines under the confidence you set are dropped, so noisy backgrounds do not produce garbage rows. Every URL is checked against its host's robots.txt first; an image without readable text is an error row, not an empty image-read row (from 3 November 2026 it costs the small image-checked fee, see Pricing).

Who this is for

Paste image URLs in urls (up to 200 per run; PNG, JPEG, WebP, GIF, BMP, TIFF) and set minConfidence (default 0.5). Every row carries the URL, the full text, the lines, the line count, the average confidence, the boxes, the image width and height, the format, the file size and the OCR time.

Who this is not for

This is OCR, not vision: printed and clean screen text reads reliably, handwriting, tiny text, heavy skew or low contrast read poorly or not at all, and the confidence column tells you which lines to trust. The models cover Latin scripts and Chinese; other scripts are not supported. Images longer than 2,600 pixels on a side are scaled down before reading; files above 20 MB are refused for free. Reading order is top to bottom, then left to right, which can interleave columns in multi-column layouts.

Input fields

Field	Type	Required or default	What it does
urls	list of text	required	One URL per row, up to 200: PNG, JPEG, WebP, GIF, BMP, TIFF, or a scanned PDF (up to 50 pages; each page counts as one image read per started 20 lines, at least one).
minConfidence	text	0.5	Lines below this OCR confidence (0 to 1) are dropped; 0.5 keeps almost everything, 0.8 only clean text.

Input example

{
    "urls": [
        "https://raw.githubusercontent.com/madmaze/pytesseract/master/tests/data/test.png"
    ],
    "minConfidence": "0.5"
}

Output example

Field	Type	What it holds	Example
url	text	The page or file this row is about.	https://raw.githubusercontent.com/madmaze/pytesseract/mas...
text	text	All text found in the image, in reading order.	This is a lot of 12 point text to test the ocr code and s...
lines	list	The same text split into lines, each with its own reading.	["This is a lot of 12 point text to test the", "ocr code...
lineCount	number	How many lines of text were found.	8
avgConfidence	number	How sure the reader is on average, from 0 to 1; below roughly 0.6 the reading deserves a check.	0.984
boxes	list	The place of every line in the image, as coordinates, so you can point back to where it stood.	["36,90,546,30 0.967", "34,123,584,33 0.994"]
width	number	The width of the image in pixels.	640
height	number	The height of the image in pixels.	480
format	text	The image format, for example png or jpeg.	PNG
fileBytes	number	The size of the file in bytes.	11513
processedMs	number	How long the reading took, in milliseconds.	3752
pageCount	number
chargedReads	number
pages	list

Error codes: INVALID_URL, ROBOTS_DISALLOWED, NOT_AN_IMAGE, TOO_LARGE, TOO_SLOW (download not finished within 20 s), NO_TEXT, FETCH_FAILED, BLOCKED.

One delivered row looks like this:

{
  "url": "https://raw.githubusercontent.com/madmaze/pytesseract/master/tests/data/test.png",
  "text": "This is a lot of 12 point text to test the\nocr code and see if it works on all types\nof file format.\nThe quick brown dog jumped over the\nlazy fox. The quick bro...",
  "lines": [
    "This is a lot of 12 point text to test the",
    "ocr code and see if it works on all types",
    "of file format.",
    "The quick brown dog jumped over the",
    "lazy fox. The quick brown dog jumped",
    "over the lazy fox. The quick brown dog",
    "jumped over the lazy fox. The quick",
    "brown dog jumped over the lazy fox."
  ],
  "lineCount": 8,
  "avgConfidence": 0.984,
  "boxes": [
    "36,90,546,30 0.967",
    "34,123,584,33 0.994",
    "37,160,189,26 0.972",
    "34,190,553,36 0.999",
    "33,225,555,34 0.981",
    "34,260,565,33 0.984",
    "41,293,523,34 0.992",
    "35,329,528,32 0.986"
  ],
  "width": 640,
  "height": 480,
  "format": "PNG",
  "fileBytes": 11513,
  "processedMs": 4251,
  "status": "ok"
}

Related actors from steadydata

• pdf-text-extractor: the same for PDF files with a text layer
• webpage-to-markdown: the text of web pages without OCR
• google-images: find the images to read

Pricing

Pay per event: one image-read event per delivered result. No charge for inputs that fail, no separate platform-usage surcharge.

From 3 November 2026 two small events are added: image-checked ($0.001) for every image that was downloaded and read to the end, also when it holds no readable text, and Apify's apify-actor-start ($0.0005 per GB, so $0.001 per run at the fixed 2 GB). Images that cannot be downloaded, are blocked, too large or too slow stay free. image-read stays as it is.

Free Apify plan: this actor delivers up to 25 rows per run for accounts on the Apify free plan, and then stops with a message. That limit is set by us, not by Apify. It exists so the actor keeps paying for itself for the people who do pay. Any paid Apify plan runs it at full size, billed per delivered row, with failed rows never charged.

Reviews: if this actor saves you time, a short review on this page is the one thing that helps most. Ratings are what other buyers look at first, and we have no other way to ask.

Use with AI agents

AI assistants such as Claude, ChatGPT or Cursor can run this actor through Apify's own MCP server. Ask for the actor steadydata/image-ocr-text by its full name, so the assistant does not pick a different one, and start from this minimal input:

{
  "urls": [
    "https://raw.githubusercontent.com/madmaze/pytesseract/master/tests/data/test.png",
    "https://raw.githubusercontent.com/madmaze/pytesseract/master/tests/data/test-european.jpg",
    "https://raw.githubusercontent.com/madmaze/pytesseract/master/tests/data/test.jpg"
  ]
}

The assistant can read the price per event from the Pricing section above before the first paid run. Failed input comes back as an error row, as described above.

FAQ

Does it read receipts and invoices? Yes, printed ones read well; use the boxes to keep amounts next to their labels.

Can I OCR a PDF? Yes, a scanned PDF by URL works like an image: every page is rendered and read (at most 50 pages per file), and the file comes back as one row with pages (text per page) and pageCount. Each page is charged as one image-read per started 20 lines of text, at least one per page (a page of 18 lines is one, a dense page of 70 lines is four), because reading time grows with the number of lines; chargedReads on the row shows the count. If the first two pages hold no readable text, reading stops there and the file is reported as NO_TEXT. A PDF that already has a text layer is cheaper and exact with the PDF Text Extractor below.

Which languages? Latin-script languages (English, Dutch, German, French, Spanish and so on) and Chinese. Cyrillic, Arabic and Asian scripts other than Chinese are not supported.

Is my image sent to a third party? No. The OCR runs inside the actor; nothing leaves Apify.

Do I pay for an image without text? Until 2 November 2026, no. From 3 November 2026 an image that was read to the end without readable text costs $0.001 (image-checked); reading it uses the same computing time as an image with text.

Is personal data collected? Only what is in the images you send: the actor returns the text it reads in them and keeps nothing after the run. Send only images you are allowed to process.

What do the fields mean? boxes entries read x,y,width,height confidence in pixels of the original image; avgConfidence is the mean over the delivered lines; processedMs is the OCR time only.

What happens when the source changes? Sources change from time to time; that is the nature of this work. The actor is monitored daily and fixed fast, and while it is broken you are not charged, because only delivered results cost anything.