Put translated text back into a scan in its original position

Steps

1. Prepare a bilingual tableGenerate one with "AI-translate whole document…", or arrange a bilingual table exported from your CAT tool with the source in column A and the translation in column B.
2. Set a glossary (optional)For AI translation, provide a "source => translation" glossary. Glossary entries are used as they are and are not sent.
3. Review the translationsCheck each entry in the keyword table edit window. Rows that contain only numbers, reference codes or currency codes are not translated.
4. Set overflow handlingTranslations are usually longer than the source. After the keyword table is generated, "When image text is too long" switches automatically to "Stay within original area", reducing the font size if needed.
5. Find, preview, saveAfter finding, check the layout in "Preview replacement", then replace and save as a new file. Read-back verification runs as usual.

What is sent out during AI translation

When you use a cloud platform, the text to be translated is sent to the platform you chose:

• Confirmation before sending: the first time text is sent to a given cloud platform, you are asked, and nothing is sent until you agree.
• Content protection (on by default): numbers, ID numbers, phone numbers, email addresses and URLs are replaced with placeholders such as ⟦1⟧ ⟦2⟧ before sending, and restored unchanged when the translation comes back. Amounts, dates and reference numbers are not altered, and these numbers are not sent. Names, addresses and other ordinary words are sent along with the text, so for files containing that kind of information, use a local model or do not use AI translation. Entries with missing or extra placeholders are not used.
• Local models: with Ollama or LM Studio, the text does not leave your computer.

For the full description of network use, see Network behavior.

Layout tips

• When the target language is much longer than the source (for example, Chinese translated into Thai or English), keep “Stay within original area” and check in the preview that the font size is not too small.
• Short terms in table cells work well for backfilling. When whole paragraphs differ a lot in length, the font size may shrink noticeably after backfilling; consider backfilling only headings and labels.

FAQ

Which AI translation platforms are supported?

DeepSeek, ChatGPT, Gemini, Grok, Claude, Qwen (Tongyi Qianwen), Kimi and Zhipu GLM, local models such as Ollama and LM Studio, and any OpenAI-compatible API.

Where are API keys stored?

Only on your computer, encrypted for the current Windows account (DPAPI). They are not written to settings files, projects or the operation record.

Am I charged again for the same sentence?

No. Sentences already translated with the same platform, model, target language and glossary are cached on your computer and are not requested again.