Document translation
Import a file, read it alongside its translation, edit passages, and download the result.
Open your file
Choose Document translation in the extension menu, then drop a file onto the page or use the file picker. Without a file, expand No file yet? Try a sample. Samples also wait for you to start translation.
For an online PDF, choose Open PDF reader from the extension menu or paste its direct link on the document start page. FluentRead opens the original in its own reader. If a link requires a login and cannot be opened, download the file first and import it.
Supported formats are PDF, ePub, Word DOCX, HTML, TXT, Markdown, JSON, and SRT, VTT, ASS, SSA, or LRC subtitles. The limit is 50 MB for a PDF and 10 MB for other formats. Convert older .doc files to .docx first.
Translate and read
- Choose the translation service, target language, and a model for AI services in the toolbar. Set the source language and glossary in the sidebar; narrow windows use Translation settings.
- Check the file, then start translation.
- Switch between original, bilingual, and translated views. Translation starts near your current reading position and completed passages appear as they arrive.
Use Files in the sidebar to switch documents and Outline to jump to chapters. Focus reading hides surrounding controls; Esc exits. Zoom with the toolbar, a trackpad pinch, or Ctrl with the mouse wheel. After clicking the document, Ctrl/Cmd +, −, and 0 also work. Zoom affects only the reading preview.
Pause, retry, or change services
Pause translation preserves completed translations and edits. Continue, or choose Retry after a failure, to process unfinished passages.
Temporary outages and rate limits show a reason and retry automatically; you can pause while waiting. To change a service or model, pause, make the change, and continue. Changing the active service or model from another page also pauses the old task. Existing results remain and the new service handles the remainder.
Changing languages, glossary settings, or some connection parameters requires a fresh translation. Confirming Retranslate replaces existing translations and manual edits, so download the current version first if needed. Expand Details to inspect an error.
Edit and download
- Open proofreading, search for a sentence or filter untranslated passages, and edit the translation directly.
- Open Download, choose Bilingual or Translation only, then use the download button below. Your edits are included.
The selected version is independent of the reading view. Partial downloads require confirmation. Passages without a translation, including cleared translations, retain their source text. Changing settings alone does not retranslate the file; downloads use the results currently kept.
The panel opens beside the toolbar. PDF versions include a small layout illustration. Expand PDF layout to choose Original or Reflow without changing the reading view. If generation fails, your translation and edits are kept. Use Retry download, or expand Technical details to inspect the cause.
Download original saves the opened source file without translating it first. You can cancel and retry ePub, DOCX, PDF, or batch ZIP generation without losing translations or edits.
Continue after a refresh
Documents, translations, and edits are stored in this browser. Continue after refreshing or restore an earlier file from Recent translations. Clearing browser data can remove them; download files you want to keep or share.
PDF limits
PDFs default to Original layout. Bilingual pages appear side by side, translated paragraphs follow the source positions, and formulas, figures, and numeric table cells keep their original appearance. Zoom scales the whole page, preserving line breaks and paragraph positions. Hover over a long passage or use the book button beside it to read the full paragraph. Choose Reflowed text for complete paragraphs in a reading layout. The toolbar offers search across both texts and translation styling.
Proofreading, full-passage comparisons, outline titles, and search terms each follow their own text direction, while the source and translation columns keep their order.
Scanned PDFs and scanned pages within text PDFs are supported. Starting translation first recognizes text from the current page, then translates it. Missing recognition packs download automatically, so check the source language first. After recognition, you can select and copy text over the source image. Text positions are estimated from recognition; use the image when checking accuracy. Recognition, formulas, and technical terms can still be wrong; review important passages.
For scanned papers, recognition uses higher-resolution page images and automatic column detection. Completed pages are saved as recognition proceeds. After pausing or a failure, continuation recognizes only the remaining scanned pages and keeps existing translations and edits. Pages with no recognized text remain available for retry and do not count as fully translated.
Numeric superscripts and subscripts are checked using enlarged crops and their positions on the original page. Confirmed scientific notation and numeric subscripts are kept unchanged during translation. If a candidate cannot be confirmed, Check original opens the source page with dashed markers. This check does not guarantee that every formula or symbol is correct.
After recognition, choose Correct recognized text in document settings to compare each line with the scan. You can also open the matching passage from the translation editor. Saving clears only the translations whose source text changed; previous translations remain available in saved revisions. Restore recognition undoes the source correction and keeps the manual source text in saved revisions.
Choose Recognize this page again to reread only the current page without starting translation. Matching manual corrections are kept; edits that cannot be matched remain in saved revisions. The original PDF file stays unchanged.
Numbers with low word confidence but complete matching glyph evidence retain their original reading with a review marker, instead of being silently dropped. These candidates can still contain misread or missing symbols; compare important values with the source image.
Low-confidence Chinese characters receive two context reads at smaller scales. Automatic correction requires matching high-confidence results and changes confined to the original low-confidence Chinese character positions. Numbers, punctuation, and other text are kept. Unconfirmed readings remain marked for review; low-quality scans can still need manual correction. Small letter scripts misread as trademark and quote clusters are marked for review when their position provides evidence; they are not guessed as digits.
Scanned list bullets are checked against their original shape and repeated indentation. Confirmed items are translated separately, and selecting the source keeps the bullets. Uncertain markers remain flagged for review; ordinary letters are not automatically treated as list symbols.
Scanned diagrams stay as original images when their boundaries can be identified; captions remain translatable. Tables with clear rules and text cells receive a separate recognition pass to fill omissions. Borderless tables with a separate caption, aligned columns, and repeated numeric columns are also organized into cells and read after the preceding body columns. No rules are drawn on the source image. Text is translated within each cell, while numeric cells stay unchanged. Conflicting readings keep the first result and show a review marker. Full-passage controls and continuation pages keep long translations available.
If extracted PDF text contains anomalous characters, a notice asks you to check it against the original. The notice does not repair characters or change the source or translation.
You can select, copy, and translate original-page text. Selection uses your separate selection translation service; the whole document uses the document service. In the browser's own PDF viewer, select text and choose Translate selected text from the context menu to open a separate result window. Enable selection translation and its context-menu entry first. The limit is 4096 characters per selection.
PDF exports, long passages, and retained edits
Original-layout downloads keep the source-page order. Bilingual output places each source page beside its translation; translation-only output keeps translated pages. Long passages retain a full PDF comment and continue on extra text pages after their matching source page, so the exported file can have more pages. Translated text can be searched, selected, and copied.
Downloading one document in Reflowed text uses fixed-size text with automatic page breaks. Bilingual output follows each source page with its translation. Figures and formulas stay together where possible; use Original layout when exact original glyphs matter. The first reflowed export prepares a font of about 5 MB online, then reuses it locally. Font requests contain no document text, translations, or credentials. Uncommon glyphs may depend on local fonts. Batch exports use Original layout.
Identical text painted again in place for bold or shadow effects is counted once in the analyzed source and translation input. Adjacent repeated words, different text, and partial overlaps are retained. Original-page pixels are unchanged.
Explanatory text above a formula is separated from raised integral and summation symbols so it is not skipped as part of the formula. Table boundaries and original-page glyphs remain unchanged.
Letter-spaced section names such as Abstract and References are identified with supporting body text and kept separate from nearby tables. Original letter spacing is preserved. Sparse textbook callouts with a heading and a list remain translatable text; a section heading with a vertical separator above the frame stays separate from the figure.
After a PDF is parsed again, edits are restored only when the source text and position can be identified safely. Other edits remain available to read and copy under retained revisions in proofreading. Clearly connected page fragments offer an optional cross-page context view; source passages and edits are not merged automatically.
Other formats, terminology, and AI context references
Markdown, HTML, and ePub aim to retain links, emphasis, and inline code. URLs, email addresses, code blocks, and program identifiers remain unchanged. If formatting cannot be preserved, retry or change services. JSON keeps its structure and non-text values; nesting beyond 1000 levels is unsupported. Word tables remain tables, and ePub downloads retain chapters, images, links, and styles.
Subtitle exports retain numbering, timing, and formatting tags. Translation-only output replaces the cue text; bilingual output keeps both texts in each cue. Fixed-width wrapped text is joined into paragraphs, while headings and lists stay separate where possible.
Translate multiple passages together is enabled by default to reduce small requests. Disable it in the current document's settings without changing webpage settings. Format protection does not guarantee accurate or complete translation. Use a glossary for consistent terminology and review key passages.
Context references
With AI enhanced translation enabled for an AI service that supports general prompts, translation can use current-chapter definitions, nearby source text, and completed translations. Edits confirmed in the current reading session take priority over machine references; your glossary has the highest priority.
This mode processes passages one by one and can be slower. References cover a limited range and cannot guarantee that every ambiguity is resolved. Maintaining context makes no extra model calls and does not persist across documents.
Save a material and resume reading
Save a material and resume reading
Choose Add to learning materials to keep the document with your saved expressions, questions, and notes. For an existing material, update the reading position. Continuing from the Learning Center returns to the saved passage. Materials and positions stay in this browser.
If the local source has been cleared, select the same file again. A different file with the same name does not overwrite the old material. If the reader cannot identify the passage, it keeps the old location and asks you to save a new one. Older materials may restore only a page or chapter, so check the position. Scrolling manually stops automatic positioning.
Batch translation
Batch translation
Drop several files together or use Files → Add files in the sidebar. Confirm the service and language, then translate the remaining files in order. One file's failure does not stop the others; Pause all preserves completed work.
Once the queue stops, read, edit, or download individual files. Download completed files (ZIP) includes completed documents only, with separate numbered folders for duplicate names. Changed settings require a separate restart confirmation for each partially translated file. Removing undownloaded results also asks for confirmation.
Translate in Obsidian
Translate in Obsidian
Follow the directory instructions to build and install the experimental Obsidian desktop plugin. Translate Markdown notes or text-based PDFs from the command palette or file context menu. It saves a sibling bilingual Markdown note without overwriting the source. This plugin currently sends text to Microsoft Translator; PDF output does not retain the original layout.
Does the file leave my computer?
The browser parses files locally. Text to translate goes to your selected service. With a cloud service, the relevant text leaves your device. See Data & privacy.