What the Word file contains
The document is a standard Office Open XML file (.docx) that the tool writes itself, on an A4 page with 2 cm margins and the Arial font. Each block of text the OCR finds becomes one Word paragraph. By default the lines inside a paragraph are kept as they were in the photo, with line breaks, which suits receipts, addresses and lists; tick "Join the lines" to get flowing paragraphs that rewrap when you change the font or the margins. When you convert several photos, each one starts on a new page under a heading with its file name. With "Add each original picture" on, the photo follows its text on a page of its own, scaled to fit, so you can compare the two. We checked the files by opening them with LibreOffice and python-docx and reading the text back.
Arabic documents in Word
Paragraphs whose first letter is Arabic are marked as right-to-left in the Word file, so they start at the right margin, and the Arabic words are written as right-to-left runs while English names, email addresses and numbers inside them stay as left-to-right runs. That keeps a line such as an Arabic label followed by an email address in the right order when you edit it. Arabic OCR is less accurate than English: across our Arabic test images about one word in ten needed a fix, mostly letters that differ only by dots, such as ت and ن. Read the text in the box before you download, or correct it in Word afterwards. The full measurements, photo by photo, are on the Image to Text page.
Several photos, one document
You can select up to 20 pictures at once, for example every page of a printed contract photographed with your phone. They are read in the order you chose them and placed in that order in the document. On a computer with four or more cores two pictures are read at the same time; in our test two photos took 3 to 4 seconds once the engine was ready. iPhone HEIC photos are decoded in the browser first. A scanned PDF works too: each page becomes a section, and you can type a page range to read only part of it. If one picture cannot be opened, the others are still converted and the problem file is named.
What is not copied
OCR reads words, not design. The Word file does not keep the original fonts, sizes, colours, columns, tables, logos, stamps or signatures; tables come out as lines of words separated by spaces, and two columns become separate blocks. Handwriting is not supported in any useful way. If your file is a digital PDF made from Word or a website, it already contains real text, and PDF to Word gives a much closer copy of the layout. If you only need the words to paste somewhere, Image to Text gives you the plain text without a document.
Privacy, download size and speed
Your photos are read by the OCR engine in this browser tab and the Word file is built on your device; neither the pictures nor the text are uploaded. The first time, the browser downloads the engine and the language model from the public jsDelivr network: about 4.4 MB for English, 3.1 MB for Arabic, or 6.0 MB for both. They are cached, so later visits start straight away. After that, a typical photo is read in 1 to 2.5 seconds on a recent computer, slower on older phones. Large photos are reduced to 3,500 pixels on the long side before reading, which keeps memory use reasonable without losing normal-sized text. If you only need the plain text, image to text is enough, and a PDF that already contains text converts better with PDF to Word.