Translation help

You can upload DOCX, PDF, EPUB, XLSX, PPTX, or TXT content. You can also paste plain text. DOCX, EPUB, XLSX, and PPTX can retain their document structure when format preservation is enabled. A format-preserving PDF is reconstructed as a DOCX because PDF is a presentation format.

The public upload limit is 250 MB per file; the transport layer allows a small amount of additional headroom. The workspace accepts plain text up to 4,000 characters.

The service extracts structure such as headings, paragraphs, tables, lists, emphasis, and supported embedded media, then rebuilds the output around the translated text. Preservation is format- and structure-dependent. Complex Office features, unusual layouts, and reconstructed PDFs should be reviewed in the final file.

Yes. The language fields accept free text as well as suggestions. The source language may be detected when supported by the workflow; always check the selected direction before starting a paid job.

Tone changes the instructions given to the translation model, for example formal, academic, technical, conversational, literary, or poetic. Tone is separate from the quality mode: it guides voice and register, while the mode controls the speed/fidelity/fluency trade-off.

Balanced: Everyday translation — Clear, faithful translation in one pass, using your tone and glossary. Precise: Additional accuracy review — Includes an additional AI review against the source for meaning, omissions, terminology and numbers. Natural: Idiomatic wording with a meaning check — Adapts phrasing to your tone and audience, followed by an AI check that the meaning is preserved.

Clear, faithful translation in one pass, using your tone and glossary.

Includes an additional AI review against the source for meaning, omissions, terminology and numbers.

Adapts phrasing to your tone and audience, followed by an AI check that the meaning is preserved.

Saved or manual glossary terms give the translator approved source-to-target terminology. You can reuse saved glossaries across jobs, add manual terms for one job, and receive a warning when a glossary’s target language does not match the form.

Use them for document-specific guidance such as keeping a product name unchanged, retaining a technical abbreviation, or using a particular register. They supplement the selected tone and glossary; they do not guarantee that every ambiguous instruction can be followed.

Before an estimate, the service performs a local structural scan. It finds embedded images in supported document packages and identifies PDF pages that have very little selectable text but contain visual content. This preflight step does not OCR the image and does not send image text to an AI provider.

The paid image path sends each selected image to Vision OCR, which returns readable source text and normalized text regions. The selected translation mode translates those regions. The service then locally edits the retained raster image: it covers a detected region and draws the translated text back into it. The image object itself stays in the document.

Images add Vision input tokens, so the estimate and processing time can increase. Image-heavy documents require an acknowledgement. Text on photographic or non-uniform backgrounds, unsupported image encodings, invalid regions, or translations that cannot fit the original region may be left unchanged or stop the job safely instead of producing a misleading replacement.

The estimate calculates source tokens and prices for the selected mode; when images are detected, it can distinguish image-inclusive and image-free scenarios. Preview translates one page or supported preview unit before payment and uses the structured translation path where available, including paragraphs, tables, and images, with a text fallback for plain content.

A job records its source and target languages, tone, quality level, formatting choice, status, pricing, and processing details. It begins awaiting payment, can expire if payment is not completed within 30 minutes, then moves through payment confirmation and processing. Completed files are available from Previous jobs; failed or cancelled jobs show their terminal state rather than generic progress.

When a job reaches a terminal state, the application clears source text, translation instructions and saved checkpoints from its active database records, and attempts to delete the temporary source file. Scheduled cleanup checks terminal jobs again, so file removal can be delayed if cleanup fails. Unpaid jobs are configured to expire after 30 minutes; expiry cleanup runs periodically, so this is not a 30-minute source-file deletion guarantee. A successful translated result is configured to remain available in Previous jobs for 30 days from completion. You can request its removal earlier using Delete result in your job history. The expiry time controls availability; scheduled file cleanup may occur later. Job records use a separate 30-day age threshold and may remain longer while a result is within its retention period or file cleanup fails. Activity records use a 90-day default with daily cleanup. Support conversations use a 30-day default with hourly cleanup. These are application defaults and schedules, not guarantees about deployed overrides. Backup copies and provider-side retention are not established by this application cleanup code.

Yes. AI translation can misunderstand context, names, terminology, or visual text, and formatting reconstruction can differ from the source. Review every result before publication, legal or medical use, or another consequential decision. Preview is useful for checking direction and style, but it is not a certification of the full document.