How to Translate a Photo of a Document (JPG/PNG)

A photo of a document — a JPG in your camera roll, a screenshot, a picture someone forwarded over WhatsApp — can be translated directly, without retyping anything. Upload the image file to a document translator that accepts images: OCR reads the text out of the picture, the text is translated, and the good tools give you back an actual translated document rather than a blob of text. Pasting the photo into a chat translator or an AI assistant also works, but it gets you raw text at best — the document itself doesn't survive.

This guide is about image files you already have, wherever they came from. It covers what actually happens when you paste a photo into a translator, why the image itself decides most of the output quality, and how to get a real translated PDF back instead of a transcript.

What pasting a photo into a translator really gets you

Plenty of tools can now look at an image: translation apps accept photo uploads, and AI chatbots happily read a picture you paste into the chat. What they return is the text — translated, often reasonably well, but as plain writing in a chat window. Everything that made the image a document is gone. A form turns into a list of disconnected labels and values, a table flattens into run-on lines where you can no longer tell which number belonged to which row, and stamps, signatures, and seals are simply not part of the output.

For a short note where you just need the gist, that's fine — paste it and read. It stops being fine the moment you need the document as a document: to keep for your records, forward to an office, print, or compare field-by-field against the original. There's a second, quieter problem too: when the output is detached from the image, OCR misreadings are invisible. A digit swapped in an amount or a mangled name looks perfectly plausible in plain text, and you have nothing lined up next to it to catch the error.

The image decides the OCR quality

Translation quality gets the attention, but with photos the bottleneck is almost always recognition. OCR is reading pixels, so everything about how those pixels were captured matters:

• Resolution — the text needs to be genuinely legible when you zoom in. If you can't read a word at full zoom, OCR can't either.
• Focus and motion blur — a slightly soft photo that looks acceptable to the eye is often enough to turn characters into guesses.
• Lighting — glare from a flash and hard shadows across the page both erase text locally.
• Angle and perspective — a page shot from the side becomes a trapezoid with distorted characters; straight-on captures read far better.
• Compression — every re-save and re-send can soften text edges further.
• Handwriting — printed text recognizes well; heavy handwriting is unreliable for any OCR system, no exceptions.

Two practical notes on where photos come from. Messaging apps commonly recompress images to save bandwidth — a document photo that has been forwarded through a few chats can arrive noticeably degraded, so if you can, ask the sender to share the original as a file or document attachment rather than a photo. Screenshots, on the other hand, are usually excellent OCR input: perfectly flat, perfectly lit, pixel-sharp. If the 'photo' you have is a screenshot of a digital document, expect good results.

From image file to translated document

To get back a document rather than a transcript, the tool has to do more than read the image — it has to rebuild the page. That's what Reglyph is built for: you upload the JPG or PNG (PDFs work too), and OCR reads the page, the original text is erased from the page image, and the translation is typeset back in its place. Tables keep their cells, numbers stay in their columns, and stamps, seals, and figures remain exactly where they were — what you download is a translated PDF that still looks like the document in the photo.

It also addresses the invisible-error problem from earlier: a bilingual side-by-side export puts the original and the translation next to each other, so names, dates, and amounts can be checked line against line instead of on faith. It supports 12+ languages, runs in any browser including on your phone, and the first 5 pages are free with no credit card — paid use starts at $5. For a single forwarded document, the free tier usually covers it entirely.

Translating multiple photos of a multi-page document

If someone photographed a contract page by page, you're holding five JPGs that are really one document. The tidiest approach is to combine them into a single PDF before translating, so the pages stay in order and you get one output file instead of five. Your phone or computer already does this: photo scan features and 'print to PDF' both turn a set of images into one PDF in seconds. Check the page order before uploading — it's the one thing no tool can fix for you afterwards.

Know what a photo translation is good for

Machine translation of a decent photo gives you an accurate, readable document for understanding, records, and everyday communication. Two limits are worth respecting. First, anything heavily handwritten — margin notes, filled-in blanks, a fully handwritten letter — should be treated as unverified until you've checked it against the original, because handwriting recognition is unreliable everywhere. Second, if the translation is destined for an authority that requires certification (immigration offices, courts), a certified human translator must review and sign it — the machine output serves as the draft and takes care of the layout, which is most of the labor, but the signature has to be human.