Extract text from a photo or image (OCR)

A photo of a book page, a screenshot containing text, a sign photographed while travelling: this tool turns any image containing printed text into text you can actually use — copyable into an email, editable in a word processor, or searchable in a PDF reader. Seven languages are supported, including Yoruba, a rare feature among free OCR tools.

What happens between uploading the photo and getting a result

Before any character recognition happens, the image goes through several automatic preparation steps, invisible to you but decisive for result quality. Text tilt is measured and corrected: a photo taken slightly crooked, which happens almost every time with a handheld shot, is straightened before going any further. If the whole page is upside down or rotated a quarter turn — a common mistake when photographing a book — the tool detects it and rights it. The image is then converted to greyscale, lightly denoised, then thresholded to sharply separate text from its background. If the characters look too small for reliable recognition, the image is automatically enlarged before the final step.

Why photo quality matters more than the engine

Optical character recognition has improved enormously in recent years, but it remains fundamentally limited by what it's given: even the best engine can't guess a letter that's completely blurred or drowned in a glare. That's why this tool shows an average confidence score after processing, rather than claiming universal reliability. Below 60%, a concrete message points you toward the most useful fixes: retake the photo perfectly flat (a book held open at an angle warps the lines of text toward the edges), with front-on rather than raking light (to avoid cast shadows on the page), without motion blur (brace yourself or use enough light for a short exposure time).

Three results for three uses

The plain text appears directly in a box you can copy with one click — the fastest way to grab a quote or a short passage. The Word document rebuilds one paragraph per detected text block, ready to be edited or completed further. The searchable PDF keeps your original photo's exact appearance while overlaying an invisible text layer: ideal for archiving a document while keeping it searchable by keyword later, without losing its original visual look (stamp, signature, layout).

Limitations to know about

This tool recognizes printed text in a relatively simple layout: paragraphs, headings, lists. A complex table with many columns, handwriting, or a very stylised font (calligraphy, artistic lettering) will give a partial or incorrect result. An image over 50 megapixels is rejected for processing-time reasons; in that case, reduce its size first with the image-resizing tool.

Frequently asked questions

My photo is slightly crooked — does that matter?

No: the tool automatically detects text tilt and corrects it before recognition, up to a point. A handheld photo with a few degrees of tilt is usually fine; a very steep shooting angle (a page photographed sharply from the side rather than head-on) is harder to correct automatically.

Why is the result sometimes poor?

Recognition quality depends almost entirely on the source image: blurry text, a photo taken in dim light, a glare on the page, or too low a resolution all badly degrade the result, whatever recognition engine is used. An average confidence score is shown after processing; below 60%, the tool suggests retaking the photo under better conditions.

Can I combine several languages, for example French and English?

Yes: tick every language present in your document before starting recognition. The engine uses them simultaneously, which improves accuracy on a bilingual document compared to declaring only one language.

Is handwriting recognized?

No, or very poorly: this tool is designed for printed text (a book, an official document, a screenshot, a sign). Handwriting follows very different recognition rules, which this engine doesn't cover.

What happens if my photo contains a table?

Text is extracted cell by cell in the reading order detected by the engine, but the table's layout (aligned columns, borders) isn't reconstructed: for a complex table, the Word result will need manual reformatting.

Are my photos kept after processing?

No: the uploaded file and the generated documents are automatically deleted 30 minutes after processing, never logged or reused. You can also click "Delete now" as soon as you've retrieved your result.

Why can't I choose a language that's missing from the list?

Only languages whose recognition data is actually installed on the server are offered: a shorter, reliable list beats an option that would silently fail.