Fully automated B-roll workflow. Claude Code transcribes your talking-head video, decides where B-roll goes, generates clips through Higgsfield, and composites them into a finished video, all from your terminal.

Setup time: ~15 min (one time) Per video: ~20 min generation + ~10 min review

🔄 How It Works

1. Drop your raw talking-head video into raw/
2. Run one command — Claude Code transcribes it with Whisper
3. Claude reads the transcript and identifies 6–8 B-roll moments
4. It presents the plan with timestamps and prompts — you approve
5. Claude drives Higgsfield via Playwright to generate each clip
6. FFmpeg composites everything with crossfades into output/final-with-broll.mp4

⚙️ One-Time Setup

1. Install FFmpeg + Whisper

Mac with Homebrew:

brew install ffmpeg
pip3 install openai-whisper

Mac without Homebrew (no sudo):

pip3 install --user openai-whisper imageio-ffmpeg
FFMPEG_BIN=$(python3 -c "import imageio_ffmpeg; print(imageio_ffmpeg.get_ffmpeg_exe())")
mkdir -p ~/bin && ln -sf "$FFMPEG_BIN" ~/bin/ffmpeg
echo 'export PATH="$HOME/bin:$HOME/Library/Python/3.9/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc

Verify:

ffmpeg -version && whisper --help && node --version