How the `--no-whisper` Option Disables Transcription in Claude Video

The --no-whisper flag prevents the watch skill from falling back to the Whisper API when no embedded captions are found, forcing it to rely solely on existing subtitles or return frames-only output.

In the bradautomates/claude-video repository, the watch skill extracts visual frames from videos and optionally generates transcripts. The --no-whisper command-line option provides a way to disable the external transcription service entirely, which is useful when API keys are unavailable or when you want to avoid network calls.

Where --no-whisper Is Defined

The flag is implemented in skills/watch/scripts/watch.py at lines 53-56. When present on the command line, the argument parser sets args.no_whisper to True.

This boolean value acts as a safety switch throughout the transcription pipeline, specifically guarding the fallback logic that invokes external AI services.

How It Disables the Whisper Fallback

The watch skill processes transcription in two distinct stages:

1. Caption extraction – If the video has embedded subtitles, yt-dlp fetches them and parse_vtt parses the data.
2. Whisper API fallback – When no captions exist, the skill normally calls the Whisper API via Groq or OpenAI.

The --no-whisper option interrupts the second stage. After attempting to load captions, the code in watch.py evaluates the following condition:

if not transcript_segments and not args.no_whisper and video_path and meta.get("has_audio"):
    backend, api_key = load_api_key(args.whisper)
    # ... API call logic ...

When args.no_whisper is True, the condition not args.no_whisper evaluates to False, short-circuiting the block. Consequently, the script skips the load_api_key and transcribe_video calls entirely, even if the video contains an audio track.

User Feedback When Transcription Is Skipped

If captions are missing and --no-whisper is active, the script proceeds with a frames-only analysis. According to the source code around lines 77-84 in watch.py, the user receives a clear message indicating that transcription was omitted because Whisper was disabled or because no API key was configured.

Practical Usage Examples

Run the watch skill normally to try captions first, then fall back to Whisper:

watch https://youtu.be/abcdEFGH

Disable the Whisper fallback to use only existing captions or receive frames-only output:

watch https://youtu.be/abcdEFGH --no-whisper

Force a specific Whisper backend when transcription is allowed (requires API key):

watch https://youtu.be/abcdEFGH --whisper openai

Summary

• The --no-whisper flag is defined in skills/watch/scripts/watch.py (lines 53-56) and sets args.no_whisper to True.
• It short-circuits the Whisper API fallback by evaluating not args.no_whisper in the transcription guard condition.
• Caption extraction continues to work normally; only the external API call is blocked.
• When transcription is skipped, the script generates a frames-only report with an explanatory message (lines 77-84).
• The behavior is tested in tests/test_watch.py, which verifies the flag properly disables the Whisper invocation.

Frequently Asked Questions

Does --no-whisper disable caption extraction?

No. The flag only blocks the Whisper API fallback. If the video contains embedded subtitles, the skill still extracts and parses them using parse_vtt from transcribe.py.

What happens if I use --no-whisper but the video has no captions?

The script skips transcription entirely and generates a frames-only analysis. You will see a message explaining that transcription was omitted because Whisper was disabled or because no API key was configured.

Where is the --no-whisper logic tested?

The behavior is verified in tests/test_watch.py, which invokes the script with the --no-whisper flag to ensure the Whisper fallback is properly bypassed.

Can I still use a specific Whisper backend without the --no-whisper flag?

Yes. Use the --whisper option followed by the backend name (e.g., --whisper openai or --whisper groq). This requires a valid API key loaded via the load_api_key function in whisper.py.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}