1. Install the tools (one time)

brew install node ffmpeg whisper-cpp
pip3 install numpy librosa

You also need Claude Code installed and logged in.

2. Download a whisper model (one time)

Whisper turns speech into word timings for the captions.

mkdir -p ~/models
curl -L -o ~/models/ggml-large-v3.bin \
  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3.bin

ggml-large-v3.bin (~3 GB) is the most accurate, and the best choice for Hindi. For a smaller, faster download use ggml-base.bin (~150 MB) instead.

3. Add the skill to Claude Code

git clone https://github.com/<your-username>/aansh-video.git
cp -r aansh-video/aansh-video ~/.claude/skills/

4. Make a video

Open a terminal and start Claude Code:

claude

Then ask for your video, for example:

/aansh-video make a podcast-style explainer video from ~/Downloads/voice.mp3.
My whisper model is at ~/models/ggml-large-v3.bin.
Here's the script: <paste your script>

Claude will set up the project, time every word, write the scenes, check still frames, and render the MP4 into the project's renders/ folder. A few minutes of audio can take 15–30 minutes to render.

5. Ask for changes in plain words

• "Make the captions bigger"
• "Change the host's shirt to red and give him glasses"
• "When I say ₹50,000, show a money card with that number"
• "Make a smaller copy for WhatsApp"