Aiditor — Video Editor Agent Pipeline(openclaw)

AI-powered video highlight generation for educational/content video editing

First Release: Class Record Video Editor Agent (executed by OpenClaw)

System Architecture

Project Vision

The Pain Point

Creating highlight reels from raw video footage is labor-intensive:

• Manual transcription and timestamp marking
• Subjective clip selection by watching entire footage
• Tedious cutting and concatenation in video editors
• Inconsistent quality across different editors

Typical workflow: 1 hour of raw footage → 30-60 minutes of manual editing → 3-5 minute highlight

Our Solution

An automated pipeline that:

1. Transcribes speech with Whisper (word-level timestamps)
2. Segments content by complete sentences (5s silence = boundary)
3. Scores clips by content quality (keyword density, engagement signals)
4. Selects top segments to match target duration (±10s tolerance)
5. Exports ready-to-use highlight clips

Time Savings

Task	Manual	Our Pipeline
Transcription	0.5-1x video duration	0.1-0.2x video duration
Clip Selection	10-30 min	Automatic (seconds)
Export	5-15 min	Automatic (seconds)
Total	30-60 min per highlight	~5-20 min (mostly Whisper)

Time saved: 50-80% (mostly in transcription and selection)

Token Consumption

Using local Whisper (CPU/GPU) instead of API — token cost is 0.

Video Duration	Pipeline Tokens	Cost
10 min	~10K	Free
30 min	~15K	Free
60 min	~20K	Free

Note: Only costs would be compute time for Whisper, no API charges.

Roadmap

• Scenario-specific templates:
  • event_recap — Event highlights (key moments, applause, Q&A)
  • meeting_minutes — Meeting summaries (decisions, action items)
  • lecture_summary — Educational content (concepts, examples)
  • training_clip — Training videos (step-by-step instructions)
• Multi-language support (non-English transcription)
• Scene detection fallback (for non-speech videos)
• Custom scoring templates per content type
• Web UI for configuration and monitoring

Current Status

• ✅ Production-ready: Successfully processed multiple projects
• ✅ Sentence-aware segmentation: Respects natural speech breaks
• ✅ Normalized scoring: Longer clips don't automatically win
• ✅ Queue system: Multi-folder batch processing with notifications
• ✅ Heartbeat monitoring: Progress tracking per SOP

---

Quick Start (Run with OpenClaw Agent)

# Clone the repo
git clone https://github.com/enderjiang/aiditor.git
cd aiditor

# Install dependencies
pip install openai-whisper
brew install ffmpeg  # macOS

# Create a job config
# Copy jobs/00_TEMPLATE.json to jobs/my_project.json and edit the paths

# Run pipeline (via OpenClaw agent)
python pipeline.py --config job_config.json

# Or use the queue system for multi-folder processing
python queue.py --add  # Interactive job creation
python queue.py        # Run all jobs sequentially

Powered by OpenClaw

This pipeline is designed to run as an OpenClaw agent task. The pipeline:

• Follows SOP (Standard Operating Procedure) defined in SOP.txt
• Uses heartbeat for progress tracking
• Supports queue system for batch processing
• Can be triggered via OpenClaw's task system

Configuration

Field	Description
source_dir	Folder containing input videos
output_dir	Where to save processed files
target_duration	Target clip length in seconds
mode	individual (one per video) or single (combine all)
template	Scoring template (default: classrecap)

See SOP.txt for detailed documentation.