video-captions

name	video-captions
description	Use when the user asks to extract captions, subtitles, or transcribe audio/speech from a video file. Triggers on "extract captions", "transcribe video", "get subtitles", "transcrever vídeo", "legendas do vídeo", or when a video file path is provided alongside a transcription request.
disable-model-invocation	true

video-captions

Extract captions or subtitles from video files using video-captions (Whisper-based CLI).

Quick Reference

Option	Default	Description
--language LANG	en	Source language code (en, pt, es, fr, …)
--model MODEL	base	Model size: tiny, base, small, medium, large
--format FORMAT	text	Output: text, srt, vtt, json
--output FILE	stdout	Save to file instead of printing

Usage

# Plain text to stdout
video-captions video.mp4

# Portuguese audio, small model
video-captions --language pt --model small video.mp4

# SRT file output
video-captions --format srt --output captions.srt video.mp4

# WebVTT with better accuracy
video-captions --model medium --format vtt talk.mkv

Model Trade-offs

• tiny / base — fast, good for short clips or clear speech
• small / medium — better accuracy, slower (recommended for accented speech)
• large — highest accuracy, requires more RAM (~10 GB)

When to Use a Larger Model

The base model is sufficient for most clear English audio. Use small or medium when:

• Audio has background noise or music
• Speaker has a strong accent
• Language is not English

Dependency

Requires openai-whisper for Python 3.12:

pip install openai-whisper --break-system-packages