Image to Text

1. Add one image. Drag a JPG, PNG or WebP onto the upload box or pick it from your device. A run takes 1 picture.
2. Look at the two options. They appear under the file once it is added. Language offers English only. Layout starts on Automatic; switch it to Single column when the picture is a plain block of paragraphs.
3. Select Convert to TXT. The picture goes to private storage and our servers read it. The progress bar counts pages, and a picture is always one page.
4. Sign in, then download. The result page asks for a free account: Google or an emailed code, with no password. You then download the .txt file, or choose Save to my documents to keep it past the temporary window.
5. Check the result. The file opens with a line such as === Page 1 (OCR confidence 93%) ===, followed by the text. Compare anything important against your picture, especially when the score is low.

Image to text, often called OCR, matches the shapes of letters in a picture to characters. It works on printed type, not on meaning: it cannot describe a photo or name what is in it, only transcribe words that are visible.

• Good candidates. Screenshots of web pages and apps, scans or flat, evenly lit photos of printed pages, slides, signs and book pages. Dark text on a plain light background is the easiest job, and light text on a solid dark background can work too.
• Poor candidates. Handwriting, decorative fonts, text laid over a busy photo, curved or crumpled surfaces, and anything blurred, tiny or low in contrast. Other languages and non-Latin alphabets are likely to come out wrong, because the engine is set to English.
• Retyping beats OCR for a line or two. Fixing mistakes can take longer than typing a short phrase. Recognition pays off for paragraphs and whole pages.
• The output is plain text. Fonts, bold type, columns and tables are not kept. For a scanned PDF, use OCR to text, which takes PDFs as well as pictures. If a PDF already has selectable text, Extract PDF text copies it with no recognition errors at all.

• The .txt file. It is UTF-8 plain text named after your picture, so invoice.jpg gives invoice.txt. A header line sits above the text and a blank line separates paragraphs or blocks. Nothing else goes into the file.
• What the confidence figure means. It averages the engine's own scores for the words it found. A high number means the engine was sure, not that every word is right, and a picture with no readable words shows 0%. A low score also raises a warning on the result page.
• Your picture is read as uploaded. Docubox does not sharpen, enlarge or straighten it before reading. A rotation flag from a phone camera is applied, transparent areas are filled with white, and an animated picture is read from its first frame.
• White text on a transparent background disappears. Because transparent areas turn white, white text on a transparent PNG is read as white on white and comes back empty. Put it on a dark background in an image editor first.
• Upside-down pictures read as nonsense, and sideways ones often do. Docubox does not detect or fix orientation, so turn the picture upright before you upload it.
• Limits. A run takes 1 picture of up to 25 MB and 40 megapixels (width times height, in millions of pixels). A picture counts as one OCR page and one conversion. A free account has 30 OCR pages and 10 conversions a day, and when the day's OCR pages run out, new runs are refused until 00:00 UTC. Pro, at $9.99 a month, includes 2000 OCR pages a month and does not raise the per-picture limits.
• Several pictures. Run them one at a time, or build one PDF from them with Images to PDF and run that PDF through the OCR tool. Images to PDF has its own limit on how many images fit in one PDF, and the OCR tool reads a PDF of up to 30 pages. Each picture becomes one page in the text file and uses one OCR page.