VideTShrts
← All workflowsOpen source · runs locally

Accurate auto captions for Shorts, free and offline

Generate word-timed captions with Whisper and burn them into a vertical video with FFmpeg, styled for TikTok, Reels and Shorts. Works offline.

Cost
Free
Time
5 to 10 minutes per clip
Level
Beginner to intermediate (command line)
Hardware
Any recent computer; a GPU speeds up transcription.

What you need

  • Whisper: Transcription and timed subtitles
  • FFmpeg: Burn styled captions into the video

Best for: Creators who already edit their clips, Privacy-sensitive footage, Non-English speakers (Whisper is multilingual).

Steps

  1. 1

    Install

    Install FFmpeg, then Whisper.

    pip install -U openai-whisper
  2. 2

    Transcribe with short caption lines

    Word timestamps plus a short line width give the 1-line, few-words-at-a-time style used on short-form video. Use --language to skip language detection. Larger models (medium, large) are more accurate and slower.

    whisper clip.mp4 --model small --language en --word_timestamps True --max_line_width 24 --max_line_count 1 --output_format srt
  3. 3

    Optional: highlight the spoken word

    Whisper can underline each word as it is spoken in the subtitle file, a karaoke-style effect.

    whisper clip.mp4 --model small --word_timestamps True --highlight_words True --max_line_width 24 --output_format srt
  4. 4

    Fix names and terms

    Open clip.srt in any text editor and correct brand names, people and jargon. Speech models most often miss proper nouns.

  5. 5

    Burn in with a short-form style

    Bold, outlined text, bottom-center, lifted above the app buttons.

    ffmpeg -i clip.mp4 -vf "subtitles=clip.srt:force_style='FontName=Arial,FontSize=16,Bold=1,Outline=3,Shadow=0,Alignment=2,MarginV=110'" -c:a copy captioned.mp4

Questions

Which Whisper model should I use?

Start with small. Use medium or large if names or accents come out wrong and you can wait longer.

Does this work for other languages?

Yes. Whisper is multilingual; set --language to the spoken language, or leave it out to auto-detect.

Sources

Commands checked against each tool's docs on November 19, 2026.

Get new workflows by email

One short email a week: new video models and tools, free-plan and price changes, and what we found worth using.

No spam. Reply "unsubscribe" to any email and you're off the list.