> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neolli.jocoding.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Translation

> Generate natural-sounding captions in 30 languages from your source transcript.

## Overview

Translation takes your transcribed captions and produces natural, spoken-register captions in your target languages. Unlike literal machine translation, Neolli's pipeline is specifically tuned for video captions — it accounts for sentence boundaries, speaker context, timing constraints, and language-specific patterns.

## Adding languages

1. From the video workspace, click **Add Languages**
2. Select target languages from the grid, or use a **preset** (Top 3, Top 5, Top 10)
3. Choose whether to generate **captions**, **dubbing**, or **both**
4. Click **Start**

<Frame>
  <img src="https://mintcdn.com/jocodinginc/3Sgx14k0qeskrvPt/_resources/screenshots/add-languages-modal.png?fit=max&auto=format&n=3Sgx14k0qeskrvPt&q=85&s=8847197525e72c948719547bd5d4828e" alt="Add Languages wizard with language grid and presets" width="2396" height="2088" data-path="_resources/screenshots/add-languages-modal.png" />
</Frame>

<Info>
  **Language presets** select the most popular YouTube audience languages by reach. Top 3 covers \~60% of global YouTube viewers, Top 10 covers \~85%.
</Info>

## Supported languages

Neolli supports translation into **30 languages**. See [Supported Languages](/workflows/supported-languages) for the full matrix with transcription and dubbing availability.

## How translation works

Neolli uses a **3-pass translation pipeline** to balance accuracy, timing, and natural speech:

<Tabs>
  <Tab title="Caption translation">
    For caption translation, the pipeline runs three passes:

    1. **Translate** — Sentence-aware translation that groups related segments and translates them as complete thoughts, maintaining speaker context and continuity
    2. **Condense** — Shortens any segments that exceed their available display time, keeping a naturalness floor (minimum 3 words) to avoid choppy output
    3. **Polish** — Fixes language-pair-specific issues like missing subjects (Korean), passive voice overuse (Japanese), or unnatural word order

    Each pass is powered by Google Gemini, tuned for spoken-register output with language-specific guidance for all 30 languages.
  </Tab>

  <Tab title="Dubbing translation">
    For dubbing, translation uses **content windows** — complete thought units that may span multiple caption segments. This produces more natural speech for voice synthesis:

    1. **Window construction** — Groups segments by speaker, gap threshold, and sentence boundaries
    2. **Translation** — Translates entire windows as coherent units
    3. **Distribution** — Maps translated text back to individual segment timestamps

    The result is dubbed audio that flows naturally while staying synchronized with the video.
  </Tab>
</Tabs>

## Metadata translation

When you translate a video, Neolli also translates the video's **title and description** into each target language. This is especially useful when [pushing to YouTube](/workflows/youtube), where localized metadata helps international discoverability.

## Editing translations

After translation completes, you can edit any segment in the [caption editor](/workflows/caption-editor). Changes are saved as a new version — you can switch between the AI-generated original and your edits at any time.

## Credit cost

Translation is charged at **31 credits per 1,000 characters** of source text, per language. A typical 10-minute video contains roughly 7,500 characters of speech. See [Credit Costs](/billing/credit-costs) for a detailed breakdown.