manual-translator

Use python-docx to read paragraphs and table cells, send each text segment to the DeepLX translate endpoint, and write the translated text back into the document. Batch processing with intermediate saves prevents losing progress on large files.

Iterate through doc.tables, then each row and cell, translating only the cell text while leaving the table structure untouched. Saving the document after each table protects against mid-process failures.

The script accepts an API key as a command-line argument or reads it from the DEEPLX_API_KEY, DEEPL_API_KEY, or TRANSLATOR_API_KEY environment variables, checked in that order. Command-line values take priority over environment variables.

The script maps common language names to codes including EN, JA, KO, ZH, FR, DE, ES, IT, PT, and RU. Both English and Chinese language names are normalized to the correct code before sending the request.

Failures usually come from API rate limits or network timeouts on long documents. The script mitigates this by translating in batches of 50 items and saving intermediate results so completed work is never lost.

Image extraction from docx is implemented, but automatic screenshot replacement is only partially implemented and listed as a planned feature. Currently you must recapture and swap interface screenshots manually.