🎬 Auto-Story Video Cutter

This notebook automates the process of turning a long interview video (e.g., .mp4) into a concise, 2-minute story sequence using Whisper for transcription, an LLM for story selection, and ffmpeg for trimming and concatenation.

---

🧩 Features

• 🎙️ Transcribe audio directly from .mp4 to .srt subtitles using OpenAI Whisper

• 🧠 Downselect meaningful dialogue segments via GPT-4o / GPT-4o-mini

• ✂️ Automatically trim the original video to those segments

• 🎞️ Concatenate trimmed clips into one continuous short story

• 💾 All outputs saved as reproducible files:

  • output.srt → raw transcript
  • story_segments.json → LLM-selected timestamps
  • trimmed_clips/ → individual sub-clips
  • story_concat.mp4 → final 2-minute video

---

⚙️ System Requirements

You’ll need the following system-level dependencies:

Dependency	Version	Purpose
Python	≥ 3.10	Core runtime
ffmpeg	≥ 4.4	Video trimming & merging
pip / venv	latest	Environment management
CUDA toolkit	optional	For GPU-accelerated Whisper (if available)

🧰 Ubuntu / WSL install

sudo apt update
sudo apt install -y ffmpeg python3 python3-venv python3-pip

For GPU users:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

---

📦 Python Dependencies

Create a virtual environment and install:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

requirements.txt

# Core
openai>=1.12.0
whisper>=1.1.10
tqdm>=4.66.0

# Optional faster alternatives
faster-whisper>=0.10.0

# Utility
moviepy>=1.0.3
jupyter>=1.0.0

---

🧠 API Setup

Export your OpenAI API key (for GPT-4o / GPT-4o-mini):

export OPENAI_API_KEY="sk-..."

You can also use Ollama locally by swapping out the OpenAI call in the notebook.

---

🚀 Workflow

1. Place your input video in the project folder, e.g. rawfootage_mary(Interview Original).mp4

2. Run the notebook cells in order:

   1. Transcription cell → generates output.srt
   2. LLM downselect cell → generates story_segments.json
   3. Trimming cell → extracts individual .mp4 clips
   4. Concatenation cell → merges them into story_concat.mp4

3. Inspect results

   • trimmed_clips/ → check each clip
   • story_concat.mp4 → final 2-minute story reel

---

🧹 Common Issues

Symptom	Likely Cause	Fix
Error opening input file	Wrong input_video path	Use full absolute path
Invalid duration for option ss	Commas in timestamps	The notebook auto-converts now, but ensure 00:00:12.560 not 00:00:12,560
Empty clips	Used -c copy (no re-encode)	Use the provided re-encode version
Model output not JSON	LLM formatting drift	Save raw output & re-prompt with stricter JSON instructions

---

🧭 Example Outputs

output.srt
story_segments.json
trimmed_clips/
├── clip_01_00-00-12.560_00-00-16.120.mp4
├── clip_02_00-01-05.120_00-01-13.080.mp4
└── ...
story_concat.mp4

---

🧩 Extending the Notebook

• Add whisperx for speaker diarization and alignment.
• Insert a second LLM stage for scene-level b-roll suggestions.
• Add text overlays or titles using ffmpeg filters or moviepy.