Translate Books with LLMs

Source Language

Target Language

cloud_upload

Drop files to translate

Support for TXT, EPUB, SRT, DOCX, and PDF

Browse Files

auto_fix_high

Drop files to refine (already translated)

Polish an existing translation — same language in, same language out

Before launching

Bilingual Output

Include original text alongside the translation. Each chunk displays the source text followed by its translation, ideal for language learning or side-by-side comparison.

Plain Text Mode (EPUB / DOCX)

Very well suited to simple novels and plain prose where structure matters less than reading the translated text. Only paragraph text is sent to the LLM, inline styles and complex markup are dropped. Headings, lists and blockquotes are kept; images stay in the output and are regrouped at the end of their original paragraph. Faster and far more reliable on small or local LLMs that struggle with [id0] placeholders. Less ideal for technical books, illustrated guides or anything where layout carries meaning. Only applies to EPUB and DOCX files.

Text Cleanup (OCR Correction)

For poorly formatted or OCR-scanned texts. Instructs the LLM to fix: broken line breaks, double spaces, punctuation errors, and merge incorrectly split paragraphs.

Glossary

Inject a curated source-to-target term list into every chunk's prompt to keep entity translations consistent. Click Manage to create, edit or auto-extract terms.

Custom Instructions

Select custom instructions from the Custom_Instructions/ folder. Use .yaml files with translation: and/or refinement: keys to scope a preset per phase. Legacy .txt files still work and apply to both phases.

terminal Activity Log expand_more

science Sample & Compare

Run multiple LLM configurations side-by-side on a few extracts from your book — no full translation needed.

Source Language

Target Language

science

Drop a book to sample

Support for TXT, EPUB, SRT, DOCX, and PDF

Browse Files

Mode

Number of samples (N)

Max chars per sample

LLMs to compare

No results yet. Pick a file, configure columns, and click Run.

tune Provider & Defaults expand_more

AI Provider

API Endpoint

API Key

Model Selection

Naming Convention

Placeholders: {originalName}, {targetLang}, {sourceLang}, {model}, {date}, {datetime}, {ext}
Example: {originalName} ({targetLang}) ({model})_{date}.{ext}

Don't auto-pause on rate limit (HTTP 429)

By default, the translation pauses after the provider rate-limits and asks you to resume manually. Enable this to keep the translation running: it will wait for Retry-After seconds (or 60s default) and auto-resume from the last checkpoint. Useful for long novels on free-tier APIs. Click "Interrupt" anytime to cancel the auto-resume.

Parallel requests

Translate several chunks at once on cloud providers that allow concurrent requests. 1 keeps the original sequential behavior. Higher values are faster but may hit rate limits, and disable cross-chunk context for consistency. Ignored for local providers (Ollama).

Chunk size (tokens)

How much source text goes into one LLM request. Bigger chunks mean fewer requests, better flow across paragraphs and lower cost, but a higher risk that the model condenses or skips text. There is no upper limit, only these constraints: the model needs roughly 550 + (chunk size × 3) tokens of context window, and a failed chunk is retried in full. Suggested range: 450 (default, safest) to 800 for tag-heavy EPUB/DOCX, up to 1500-2000 for plain prose on a large-context cloud model. Actual chunks average about 80% of this value.

Saves API keys, endpoints, naming convention, and current model as default values in the .env file. Changes in this section are not auto-saved.

notifications_active Notifications expand_more

Get pinged on your phone, Discord channel, or any HTTP endpoint when a translation reaches a terminal state. Useful for long-running jobs with local models. Compatible with ntfy, gotify, Discord, Slack, Healthchecks.io, or any custom webhook. See docs/NOTIFICATIONS.md for the full reference.

Quick presets

Pick a service to pre-fill URL and payload, then edit the values to match your account.

Webhook URL

Leave empty to disable notifications entirely. Placeholders supported: {event}, {file}, {translation_id}, ...

Method

Headers (JSON object, optional)

For services requiring auth (Bearer tokens, API keys). Header values are also templatable.

Payload template (JSON, optional)

Leave empty for the default gotify-compatible payload. String values support placeholders like {event}, {file}, {duration_seconds}, {provider}, {model}, {source_lang}, {target_lang}, {error}.

Trigger events

Success — translation completed Failure — fatal error Interruption — paused / rate-limit

HTTP timeout (seconds)

Increase if you see TLS handshake timeouts on slow networks.

Save and test

Click Save first, then Send test. The test uses the configuration currently on disk.

folder_open

File Management

Select All

Loading files...

Total: 0 files, 0 MB

menu_book

Glossaries

Tired of seeing the same character's name translated three different ways? A glossary gives the model a fixed mapping for any term you want to control: names, places, domain vocabulary, or stylistic choices. Reusable across jobs. Assign one below (or via Settings → Translation Options) to inject it into every chunk's prompt.

Name	Source	Target	Terms	Actions

Edit glossary

Name check_circleSaved

Source check_circleSaved

Target check_circleSaved

Inflected forms: separate variants with | in the Source cell — e.g. Москва|Москве|Москвы → Moscou. Any form found in the text triggers the same translation.

	Source ⇅	Target ⇅	Category ⇅	Gender ⇅

Edits are saved automatically when you leave a field. Empty source rows are dropped on save.

brush

Styles

Give the model a reusable writing style — register, rhythm, imagery, dialogue conventions — extracted from books you already like or from a quick sample translation. Assign one below (or via Settings → Translation Options) to apply it to every chunk.

Name	Description	Phases	Mode	Actions

Edit style

Description

Narrative setting

The world the text takes place in: period, technological level, social frame. This is what keeps the translation from reaching for a word that belongs to another era. Leave empty when it does not apply.

Style rules

Translation instructions

Refinement instructions

Editing the text below directly disconnects it from the rules above. Use "Reassemble from rules" to regenerate it from the checked rules.

TBL v1.5.17

Updating...

Text translation

Drop files to translate

Drop files to refine (already translated)

Selected Files:

Before launching

monitoring Translation Progress

0

0

0

0

0s

--

$0.0000

0

description Last Translation Preview

volume_up Audio Generation

Paused Translations

Translation History

science Sample & Compare

Drop a book to sample

File Management

Glossaries

Edit glossary

Styles

Edit style

Text translation

Drop files to translate

Drop files to refine (already translated)

Selected Files:

Before launching

monitoring Translation Progress

0

0

0

0

0s

--

$0.0000

0

description Last Translation Preview

volume_up Audio Generation

Paused Translations

Translation History

science Sample & Compare

Drop a book to sample

File Management

Glossaries

Edit glossary

Styles

Edit style

auto_fix_high Extract a style from books

Candidate rules

Preview

auto_fix_high Auto-extract candidates

visibility Preview prompt block

science Quick LLM test

Extract a style from books

Auto-extract candidates