Text Translation

Text translation is the lightest entry point in Transbox: type some text, pick an engine, get the result. It lives both in the main window (Smart Translation → Quick Translation) and in the standalone mini window.

Opening Text Translation

• Main window: left menu Smart Translation → Quick Translation (opens on the Text tab)
• System tray (bottom-right of the desktop) → Text Translation
• The mini window, opened from the tray menu or by a double-press action
• Selection translation: when the selection mode is "mini window", Alt+C opens the mini window and translates the selected text

Translating

1. Choose source and target languages. The source can be Auto detect, and the detected language is shown back in the selector.
2. Choose an engine. The dropdown is grouped into three sections:
   • MT engines (free for a limited time): Baidu, Microsoft, Volcengine, NiuTrans, Alibaba, DeepL
   • Platform models (consumes credits): platform-hosted LLMs, better quality, billed by usage
   • Custom models (direct from this device): your own OpenAI-compatible model, requests sent straight from your machine. This group requires a plan that supports it and is greyed out otherwise.
3. Type or paste the source text.
4. Click Translate. The result appears in the right-hand box, where you can edit and copy it.

Input Limits

• Main window Text tab: 500 characters per request
• Mini window: 1,000 characters per request

Reuse and Terminology

• Local cache: the same engine + language pair + source text returns the cached result, with no extra request and no extra cost. Browse or clear it in Translation History.
• Term injection: when translating with an LLM engine, Transbox looks up matching terms in your local termbases (up to 10 entries whose source and target differ) and injects them into the prompt so the terminology stays consistent.

The Three Selection Modes

Choose one under Settings → Shortcuts → Selection translation behavior:

1. Mini window (default): the hotkey opens the mini window and translates the selection
2. Follow the cursor: the translation pops up next to the mouse pointer
3. Silent: nothing is displayed; press the Apply translation hotkey (default Alt+V) to insert the result at the caret — handy for writing and filling in forms

Notes:

• Silent mode uses Baidu MT, with the target language taken from Selection & screenshot target language
• Selection translation simulates a copy to grab the text, so apps running as administrator or blocking external text access will not work
• Follow and silent modes use the target language from Settings → General → Selection & screenshot target language, independent of project settings

The Mini Window

• Switch source/target languages and the engine
• Pick image (.jpg / .jpeg / .png) and Screenshot run OCR and send the text into the window
• The window can be pinned so it stays open when you switch apps
• Drag to resize

Double-press Actions

Under Settings → Shortcuts → Double-press trigger you can bind double Ctrl and double Alt to one of: disabled (default), show/hide the main window, show/hide the mini window, mini-window translation, follow translation, or silent translation of the current selection.

Tips

Efficient Workflow

1. Combine hotkeys
   • Alt+C translate the current selection
   • Alt+Z screenshot, OCR and translate
   • Alt+V in silent mode to insert the translation
2. Let the cache work
   • Repeating the same text with the same engine returns instantly
   • The cache is keyed by engine, so comparing engines never pollutes results
3. Compare engines
   • Translating the same text with another engine is the fastest way to compare; LLMs usually read more naturally on long sentences and domain text
   • For terminology-heavy content, build the termbase first and then use an LLM engine

Typical Scenarios

• Reading papers: screenshot translation for text you cannot select
• Programming: selection translation for code comments and docs
• Business email: LLM engines produce more idiomatic wording
• Filling forms: silent mode + Alt+V

FAQ

Q: Why is translation sometimes slow?

A: Network quality, the engine's server response time, and text length all matter. LLMs generate token by token, so long texts are noticeably slower than MT engines.

Q: How can I improve accuracy?

A: Match the engine to the content (MT for general text, LLM for nuanced text), fix terminology in a termbase first, and split long texts to keep context coherent.

Q: Why do I get "LLM translation does not support automatic source language detection"?

A: LLM engines need an explicit source language to build the prompt. Select it manually; if you are unsure, detect it with an MT engine first, then switch.

Q: Why doesn't selection translation work in some apps?

A: Common causes: the target app runs as administrator while Transbox does not, the app uses special rendering (games, video players), or it blocks external text access. Try running Transbox as administrator, or use screenshot translation instead.

Q: What if I run out of credits?

A: Platform models and cloud services consume credits. Switch to the MT engines (free for a limited time) group, configure your own MT API key or custom model in Settings, or top up in the user center.

Q: How do I resolve a hotkey conflict?

A: Go to Settings → Shortcuts and change or disable it. A shortcut must be one or more of Ctrl / Alt / Shift plus a letter or Enter; single function keys are not supported.

Q: The translation is too long to display. What now?

A: Both the main window and the mini window resize the result box with the window — drag to enlarge, scroll inside the box, or copy the text into an editor.

Q: Can I export translation history?

A: Not currently. History is a local translation cache used to speed up repeated translations and to look up past results; it can be cleared from that page.

Q: Can I sync translations across devices?

A: All data is local; there is no cloud sync. To move devices, export the database via Settings → Network & advanced → Data backup and import it on the new machine.