Image OCR Extractor: image to text with line and word boxes

Give it image links and get back the text in them, with a confidence score, the position of every line (and optionally every word), the image size and the orientation that was applied. It reads PNG, JPG, WebP, GIF, BMP and TIFF, cleans up low-resolution and low-contrast photos on its own, and turns sideways or upside-down images upright. OCR runs inside the actor: no API key, no external OCR or AI service.

$6 per 1,000 images, platform usage included. Images with no readable text and links that are not images cost nothing.

What you get

One row per image:

Field	Description
url	The image link
text	The recognized text, lines separated by new lines, blocks by a blank line
confidence	Average OCR confidence from 0 to 100
wordCount, lineCount	Size of what was read
lines	With the "lines" or "words" output: every line with text, confidence and a bbox (x0, y0, x1, y1)
words	With the "words" output: every word with text, confidence and bbox
width, height, format	Size in pixels (of the upright image) and the file format
orientationDegrees	How far the image was turned to read it (0, 90, 180 or 270)
preprocessing, languages	What was used
fileSizeBytes, capturedAt	File size and run time
status, errorCode, errorMessage	ok, or an error row (see below)

Box coordinates are in pixels of the upright original image, so you can draw them straight on the picture. EXIF rotation from phones is applied first.

Example row (real, from a run on 2026-10-16, shortened)

{
  "url": "https://raw.githubusercontent.com/naptha/tesseract.js/master/tests/assets/images/testocr.png",
  "status": "ok",
  "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.\n\nThe quick brown do ...",
  "confidence": 95,
  "wordCount": 60,
  "lineCount": 8,
  "languages": ["eng"],
  "format": "png",
  "width": 640,
  "height": 480,
  "orientationDegrees": 0,
  "preprocessing": "auto",
  "lines": [
    { "text": "This is a lot of 12 point text to test the", "confidence": 95, "bbox": { "x0": 36, "y0": 92, "x1": 580, "y1": 122 } }
  ]
}

Settings that change the result

• Output detail. Text only, text with line boxes (default), or text with line and word boxes.
• Languages. English, French, German, Spanish, Italian, Portuguese and Dutch. Pick up to two for mixed text.
• What the image contains. Page or screenshot (automatic layout), one block, one line, or scattered text such as signs, memes and app screenshots. Telling it saves a failed read on cropped labels and codes.
• Image clean-up. Auto (default) converts to grayscale, raises contrast, enlarges small images up to three times, and retries a weak read on a hard black-and-white copy. You can also force grayscale, a black and white threshold, or no changes at all.
• Fix sideways and upside-down images. On by default. When a read is weak or the text runs vertically, the orientation is detected and the image is read again upright.

Errors are rows, not crashes

A link that cannot be read produces an error row (status: "error") and costs nothing:

errorCode	Meaning
not-an-image	The link returned a web page or an unsupported file (the content type is named)
looks-like-pdf	The link is a PDF. Use PDF Text Extractor for those
no-text	A readable image where OCR found no text
not-found	HTTP 404 or 410
too-large	Over the 25 MB file limit
corrupt	The image is damaged or larger than 80 megapixels
fetch-failed	The server refused or timed out on every attempt

The run status message summarizes the same, and a run where every link failed ends as failed instead of pretending to succeed. A low-confidence read (under 50) is delivered with a note in the status.

Limits

• 25 MB per image, 1,000 images per run. Images over 6,000 pixels on the long side are shrunk for reading, boxes still come back in original pixels.
• Animated GIFs and multi-page TIFFs use their first frame. For multi-page documents use PDFs.
• Printed text only. Handwriting and very stylized fonts read poorly.
• Each image is downloaded directly first. If the server refuses, a datacenter proxy is tried. Files behind a login are not supported.
• Speed at the default memory: about a second for a small image, 5 to 10 seconds for a full dense page, more for photos that need turning or a second read.

Input examples

Text and line boxes for two screenshots:

{
  "urls": [
    { "url": "https://raw.githubusercontent.com/naptha/tesseract.js/master/tests/assets/images/testocr.png" },
    { "url": "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif" }
  ],
  "output": "lines"
}

Word boxes for a French and English scan, no clean-up:

{
  "urls": [{ "url": "https://example.com/menu.jpg" }],
  "output": "words",
  "languages": ["fra", "eng"],
  "preprocessing": "none"
}

A cropped single line, such as a label or a license plate photo:

{
  "urls": [{ "url": "https://example.com/label-crop.png" }],
  "imageType": "line",
  "output": "text"
}

Who uses this

• Automation and data teams. Turn screenshots, receipts and photographed forms into text for parsers and spreadsheets.
• AI and agent builders. Give a text-only model the words in an image, with positions when layout matters.
• QA and support teams. Read error screenshots and UI captures at scale.
• Archivists and researchers. Digitize scanned pages and photos of printed text.

FAQ

Can I get the position of each word? Yes. Pick the "words" output and every word has a bounding box and a confidence.

Does it work on photos taken sideways? Yes. EXIF rotation is applied and a sideways or upside-down photo is detected and read upright. The turn is reported in orientationDegrees.

What about low-resolution or blurry images? Auto clean-up enlarges small images and retries weak reads on a high-contrast copy. Very blurry text still reads poorly.

Which image formats work? PNG, JPG, WebP, GIF, BMP and TIFF.

Does it need an API key? No. OCR runs inside the actor.

What about PDFs? Use PDF Text Extractor, which also OCRs scanned pages.

Keywords

image ocr, image to text, ocr extractor, text extractor, extract text from image, screenshot to text, jpg to text, png to text, ocr bounding boxes, scanned image to text.