Soviet birth certificate, handwritten
Uzbekistan, issued 1973, Uzbek and Russian
Exactly what the product returned. Nobody fixed these files by hand. Scan: Wikimedia Commons, Public domain
A chat model reads a photo of a document well and translates it decently. What comes back is text. The tables, the stamps, the signatures and the page itself are yours to rebuild.
6,000 free characters to start, then topped up to 1,000 every month, no card. See real scans translated, with their Word files
It reads clear scans, copes with mixed languages and, with a good prompt, gives a solid first draft. For understanding a letter or translating one paragraph, a chat is often all you need.
The answer is text or Markdown, not the page. Tables are flattened or re-drawn as Markdown, stamps and signatures are described or dropped, and on a long multi-page PDF a chat can shorten or skip parts unless you go page by page. Consistent terms across a batch depend on you pasting the same instructions every time.
For a client deliverable that means someone still rebuilds the layout by hand, which is the slowest part of working with scans.
| Chat (ChatGPT, Claude) | ChatsControl | |
|---|---|---|
| What you get back | Text in the chat window | A Word file laid out like the page, translated |
| Tables | Flattened or rewritten as Markdown | Structure kept |
| Stamps and signatures | Described in words or lost | Kept as pictures where they stand |
| Editing | Ask again, copy, paste | Edit the Word file |
| Output file | You make it | An editable DOCX laid out like the scan, translated |
Compare the scan with the copy, switch to the translation, and download both files - open them in Word and check for yourself.
Uzbekistan, issued 1973, Uzbek and Russian
Exactly what the product returned. Nobody fixed these files by hand. Scan: Wikimedia Commons, Public domain
Uzbekistan, photographed on a desk
Exactly what the product returned. Nobody fixed these files by hand. Scan: Solijon Solayev, CC BY-SA 4.0
Brazil, 2021, Portuguese, English and French
Exactly what the product returned. Nobody fixed these files by hand. Scan: Civil registry office in Rio de Janeiro, Public domain
It uses language models for recognition and translation, like a chat does. The difference is everything around them: rebuilding the page as an editable Word file, keeping tables and stamps, a checker that compares the copy with the scan, a glossary and a self-check pass.
Yes. Before translating you can set the domain, audience, tone and preferred terms, and they are applied to the whole document.
A certificate, a diploma, a form with a table. You will see in two minutes whether it saves your team time.