Skip to main content

Subtitle Translator

The Subtitle Translator is an automated task designed to translate subtitle files from a source language to a target language.

By leveraging local or remote Large Language Models (LLMs), it doesn't just translate word-for-word—it reads surrounding text to preserve the narrative flow, emotional tone, and contextual nuances of your media.

Configuration Settings​

Concurrent Tasks​

  • Description: The maximum number of subtitle translation tasks that can run simultaneously.
  • Usage: If you are running a local LLM on a machine with limited hardware, keeping this at 1 or 2 ensures your system resources aren't overwhelmed.

Batch Size​

  • Description: The number of subtitle lines (cues) sent to the LLM in a single request.
  • Range: 5 to 30.
  • Usage: Larger batches speed up overall translation time but require a more capable model. For smaller systems, smaller batches prevent translation errors or text truncation.

Context Lines Before / After​

  • Description: The number of preceding and following subtitle lines provided to the LLM as background story context for each batch.
  • Range: 0 to 10 lines.
  • Why it matters: The LLM uses these extra lines purely for narrative awareness (understanding who is speaking, the setting, or the tense) but will not translate them again. This keeps your translations highly accurate to the actual story without breaking chronological order.

LLM​

  • Description: Selects the configured LLM provider instance (e.g., Ollama, LM Studio, or OpenRouter) responsible for handling the translation logic.

When running models locally (via tools like Ollama), performance is heavily tied to your available system RAM and Video RAM (VRAM). Below are recommended configuration profiles based on your hardware footprint:

System / Hardware ProfileRecommended ModelBatch SizeContext (Before/After)Concurrent Tasks
Low-End System
(8GB - 16GB RAM, No Dedicated GPU)		translategemma:4b5 - 101 - 21
Mid-Range System
(16GB+ RAM, 6GB - 8GB VRAM GPU)		translategemma:4b15 - 202 - 31 - 2
High-End System
(32GB+ RAM, 12GB+ VRAM GPU / Mac Studio)		translategemma:4b or large general model25 - 304 - 52
Cloud API Backend
(Via OpenRouter endpoints)		Select Translation-tuned Cloud Model3052+

Quick Test Tool​

The interface includes a Quick Test sandbox so you can verify your configuration and model performance before committing it to an automated library workflow.

  1. Source / Target Language Code: Enter the standard 3-letter ISO 639-2/T language codes (for example: eng for English, spa for Spanish, ara for Arabic).
  2. SRT Content: Paste a raw snippet of your subtitle track into the text area.
  3. Run Translation Test: Click this button to run a live test pass against your selected LLM. The translated output, alongside quick processing stats, will render immediately in the Translated SRT Output block.

Model Recommendation​

Model Recommendation

While this feature is compatible with various general-purpose models, language translation requires an understanding of syntax boundaries that small models often struggle with.

We highly recommend trying translategemma:4b via Ollama. Despite its compact 4-billion parameter size making it incredibly fast and lightweight for local hardware deployment, it is specifically fine-tuned and highly optimized for translation pipelines. It respects line structural constraints remarkably well where other smaller models often fail.