ClipMaker

A local web app for turning long-form interview videos into short clips — with AI transcription, speaker diarization, interactive editing, and CSV export.

Stack

Layer	Tech
Frontend	React + TypeScript + Vite + Tailwind
Backend	FastAPI (Python)
Transcription	faster-whisper
Speaker ID	pyannote.audio 3.1
Clip cutting	ffmpeg
Descriptions	OpenAI GPT-4o
YouTube download	yt-dlp

Prerequisites

• Python 3.10+ and pip
• Node.js 18+ and npm
• ffmpeg installed (brew install ffmpeg on macOS)
• yt-dlp installed (brew install yt-dlp or pip install yt-dlp)

Setup

1. Clone and configure

   cp .env.example .env
   # Edit .env with your API keys

2. API keys in .env

   • OPENAI_API_KEY — for GPT-4o clip descriptions (optional but recommended)
   • HF_TOKEN — for speaker diarization via pyannote.audio (optional)
     • Accept model terms at https://huggingface.co/pyannote/speaker-diarization-3.1

3. Start everything

   chmod +x start.sh
   ./start.sh

   Then open http://localhost:5173

Workflow

1. Upload — Paste a YouTube URL or choose a local video file. Enter interviewer and interviewee names.
2. Process — Whisper transcribes the audio; pyannote.audio identifies speakers. Takes a few minutes.
3. Transcript — Click a segment to start a selection, click another to end it. Hit Create Clip.
4. Clips — For each clip:
   • Drag the green/red markers on the timeline to fine-tune start/end times
   • Enter thumbnail text
   • Generate a description with GPT-4o (or write your own)
   • Click Cut clip to render the video segment
5. Export — Preview and download clips.csv with columns: Artist | Start | End | Thumbnail | Title | Description | Full link

Notes

• All data is stored in memory and on disk under backend/jobs/. Nothing leaves your machine except API calls to OpenAI/HuggingFace.
• Speaker diarization can be skipped (no HF token needed) — speakers will show as SPEAKER_00, SPEAKER_01 etc.
• Use WHISPER_MODEL=base for faster (less accurate) transcription, large-v3 for best results.