pyVideoTrans

Appearance

Short Video Optimization Guide ​

Short-form videos (such as TikTok, Kuaishou, YouTube Shorts, Instagram Reels) have different translation requirements compared to long-form content. Subtitles on short videos must be concise and punchy to fit compact vertical screens. Below is the recommended optimization guide for short videos.

Core Principles ​

The key to short video subtitling is brevity:

  • Fewer characters per subtitle line
  • Shorter subtitle display duration
  • Appropriately smaller font size

Optimization Steps ​

1. Select the Best STT Engine & Settings ​

We recommend using faster-whisper(Built-in) + large-v3 model and enabling SecondarySTT.

Adjust parameters in Advanced Options -> ASR Settings:

  • VAD:Max speech duration(s): Set to 3-6 (shorter durations help create shorter subtitle segments).
  • VAD: min speech duration(ms): Set to 1000-3000 (helps keep subtitle segments concise).

⚠️ Shorter duration thresholds create shorter subtitle lines but may slightly lower recognition accuracy. Longer durations improve recognition quality but produce longer subtitle sentences.

2. Control Subtitle Character Length ​

In Advanced Options -> Alignment:

  • Number of Characters per line for CJK: Set to 10-15 characters.
  • Number of Characters per line for Other: Set to 20-40 characters.

This prevents long subtitles from running off the screen.

3. Reduce Font Size ​

If subtitles still appear too long on screen, navigate to More... -> Subtitle style -> Font Size and reduce the font size by 1–2 points.

4. Remove Punctuation ​

Under More... -> Default Punct, select Remove Punct. This delivers a cleaner, more modern look for short-form video captions.

5. Use LLM Re-segmentation (Optional) ​

If you are not using the clone voice role, you can enable LLM Re-segment and configure a DeepSeek API key.

Note: When using the clone voice role, do NOT enable LLM Re-segment. Resegmenting will disrupt the original audio timestamps and degrade voice cloning quality.

6. Translation Engine ​

Select DeepSeek or another AI engine paired with a flagship online LLM model.

7. Voice Dubbing Engine ​

Select a TTS engine based on your needs. If using the clone voice role:

  • Make sure to uncheck LLM Re-segment.
  • Ensure the following in Advanced Options -> ASR Settings:
    • VAD:Max speech duration(s): Set to 3-6
    • VAD: min speech duration(ms): Set to 1000-3000

Before vs. After Comparison ​

Before OptimizationAfter Optimization
Subtitles too long, overflow screen edgesShort and compact, fits vertical screens perfectly
Font too large, obscures video contentClean font size, unobtrusive
Punctuation clutters reading flowPunctuation removed, clean and legible
Long subtitle durationFast-paced, dynamic rhythm matching short video style