Module 04 / 06

Translation Studio

A whole document translated segment by segment, source and translation side by side; every segment can be retranslated or rewritten by hand, then exported.

Module specs
Runs100% on-device translation
InputTXT / Markdown / Word / CSV
Feature 01 / 02

Document translation

Plain text, Markdown, Word and CSV import directly for segment-by-segment translation; scans and PDFs go through Document Recognition first, then import with one click.

  • Segment by segmentTranslation runs in the background and lands one segment at a time; leaving the page doesn't stop it, and coming back picks up where it was.
  • Context promptDescribe the document's field and purpose first (pharmaceutical regulation, legal contract, patent…) and the whole file is translated in that same context.
  • Structure preservedHeadings, lists, quotes and code blocks stay as they are; embedded images are not translated.
Translation Studio — TranslateSimulated demo
Source
Translation

本節說明離線推論的整體架構。

This section describes the overall architecture of on-device inference.Not translated yet

所有模型權重都保存在使用者的裝置上。

All model weights are stored on the user's own device.Not translated yet

延遲取決於硬體,而非機房負載。

Latency is determined by the hardware, not by datacenter load.Not translated yet

因此尖峰時段不會出現排隊或降速。

There is therefore no queueing or throttling at peak hours.Not translated yet

Context: technologyTranslation complete
Feature 02 / 02

Side-by-side refinement

Source on the left, translation on the right, segment by segment; anything you're not happy with can be retranslated on its own, or rewritten by hand.

  • Per-segment toolsEvery segment can be edited, retranslated or copied; anything you change by hand is marked "Edited manually".
  • Two reading modesBilingual side by side, or collapse the source and read the translation alone.
  • Failures don't blockA segment that fails to translate is flagged on its own and can be retried; the rest export as usual.
Translation Studio — RefineSimulated demo
Source
Translation

本節說明離線推論的整體架構。

This section describes the overall architecture of on-device inference.

延遲取決於硬體,而非機房負載。

Latency depends on your hardware, not on datacenter load.

Edited manually

因此尖峰時段不會出現排隊或降速。

No queueing or throttling occurs at peak hours.

Next

Got the machine? Four steps.

Install, agree, activate, download models — most of the time is just the download.