Image and PDF OCR to Text

What it does

This Actor reads the text in images and PDF pages with OCR (optical character recognition). Give it the web address of one file or a list of them. For each image, and for each page of each PDF, you get the text, one line per line, plus a confidence score from 0 to 1 and every line with its own score and its position on the page.

It reads PNG, JPEG, GIF, BMP, TIFF and WebP images, and PDFs, including scanned PDFs with no text layer. The file type is read from the file itself, not from its name.

It reads each site's robots.txt before anything else and obeys it. It never logs in, never sends a cookie and never tries to get past a CAPTCHA.

How to use it

1. Open the Actor in Apify Console.
2. Put one web address per line in Image or PDF URLs. Files you upload in Console work too.
3. Set Maximum pages per PDF if your PDFs are long.
4. Click Start.
5. When the run ends, open the Output tab. The Text view shows the text and confidence for each image or page. Export as CSV, JSON or Excel.

Input

Field	What it does	Default	Limits
Image or PDF URLs (fileUrls)	The files to read, one address per line.	None. Prefilled with one public sample image	http and https only. Duplicates are removed
Maximum files (maxFiles)	The most files this run reads.	50	1 to 1,000
Maximum pages per PDF (maxPdfPages)	The most pages read from each PDF, from page 1.	10	1 to 200
Maximum file size (maxFileSizeMb)	Larger files are not downloaded.	20 MB	1 to 50 MB
Download timeout (requestTimeoutSecs)	How long a server may take to start answering. The whole download may take up to four times this.	30 seconds	5 to 120 seconds

Output

One dataset item per image, and one per PDF page read:

Field	What it holds
sourceUrl, finalUrl	The address you gave, and where the file came from after any redirects
fileType	application/pdf, image/png, image/jpeg, image/gif, image/bmp, image/tiff or image/webp
pageNumber, pageCount	The PDF page from 1, and the pages in the PDF. For an image, empty and 1
text	All the text read, one line per line, top to bottom
confidence	0 to 1: the average line score, weighted by line length. Empty when no text was found
textFound, lineCount, characterCount	Whether any text was found, and how much
lines	Every line: its text, its confidence, and its box as four [x, y] corners in pixels
imageWidth, imageHeight	The size of the image that was read
ocrMs, processedAt	How long the OCR took, and when
note	Set when only the first frame of an animated or multi-page image was read

Example row from a local test run, a two-line PNG:

{"fileType": "image/png", "pageNumber": null, "pageCount": 1, "text": "Invoice 2026 for Daddy D Prints\nTotal due: 42.50 USD", "confidence": 0.9998, "textFound": true, "lineCount": 2}

An image or page with no readable text still gives a row, with textFound false and an empty text.

How the text is read

The OCR engine is RapidOCR on ONNX Runtime, with the PP-OCRv6 small models that ship inside the package. Nothing is downloaded while the Actor runs, and your files are not sent anywhere else. PDF pages are drawn at up to 144 dpi and then read like images. Text is found on the image scaled to at most 1,280 pixels on its longest side, which keeps a run inside 512 MB; ordinary 10 to 11 point document text on a full page reads correctly at that size. The models are trained mainly on Chinese and English printed text. Handwriting and very small print read less reliably, and the confidence score shows it.

Pricing

This Actor is priced per event, with two charges:

• Actor start: Apify's automatic start event, $0.0001 per start (one event per GB of run memory, minimum one).
• Page read: one charge for each image or PDF page written to the dataset.

Price per image or PDF page: $3.00 per 1,000, which is $0.003 each.

These are never charged: files that could not be downloaded (robots.txt blocks, timeouts, errors, 401, 403, 429, login and CAPTCHA pages), files that are not an image or a PDF, damaged files, files over the size limit, PDF pages past Maximum pages per PDF, and the REPORT and SKIPPED records.

If you set a maximum cost for the run and it is reached, the Actor reads no more pages, lists the rest in SKIPPED as spending_limit_reached, and ends the run without an error.

Records

Record	What it holds
REPORT	Why the run stopped, the settings used, files and pages read, file types, average confidence, pages with no text, skipped by reason, robots.txt results and requests per site
SKIPPED	Every file or PDF page that was not read, with the reason

Rules it follows

• robots.txt is read first on every site, including every site a redirect leads to, and obeyed.
• One request at a time to any one site, with the site's Crawl-delay. A Crawl-delay over 30 seconds skips the site.
• On Apify, private and local network addresses are refused.
• There is no cookie, header, login, proxy or user agent input. The user agent starts with image-pdf-ocr-text/0.1.

Your responsibility

You are responsible for having the right to read the files you give this Actor and for following each website's terms of use. The text read from a file is yours to handle; the Actor keeps it only in your run's storage.

Limits

Limit	Value
Files per run	1 to 1,000
Pages per PDF	1 to 200
File size	1 to 50 MB
Image size	Up to 100 million pixels; larger images are scaled down to 4,000 pixels on the longest side before reading
Password-protected or damaged PDFs	Skipped, listed in SKIPPED
Memory	512 MB by default