AutoClipWeb is an automated AI-powered video clipping engine. It transforms horizontal YouTube videos (16:9) into viral vertical clips (9:16) for TikTok, Reels, and Shorts.
It leverages OpenCV (YuNet) for intelligent speaker tracking, Faster-Whisper for high-quality SubStation Alpha (ASS) subtitles, and Gemini Flash 3.6 for analyzing and selecting the most engaging, viral moments.
- Frontend Performance: Statically built Tailwind CSS and highly optimized vanilla JS DOM updates.
- Relational DB: SQLite with SQLAlchemy Async, segmented models (VideoJobs, Clips), managed via Alembic.
- Hardware Acceleration: Automatic NVENC encoding fallback, substantially speeding up FFmpeg render times on supported GPUs.
- TikTok Integration: Secure, robust OAuth pipeline with auto token-refresh mechanism for direct application-to-TikTok publishing.
- Security & Stability: Strong file path traversal defenses (
validate_export_path) and active task cancellation/garbage collection.
-
Prerequisites
- Python 3.13
- FFmpeg installed and accessible in your system
PATH. - SQLite
-
Install Dependencies
pip install -r requirements.txt
-
Database Migration Initialize the database to the latest schema using Alembic:
alembic upgrade head
-
Run the Application
python -m uvicorn main:app --port 8000 --reload
Access the web interface at
http://127.0.0.1:8000/.
For deep technical insight into the AI algorithms (Rule of Thirds framing, Hallucination-filtering, Garbage collection), refer to SYSTEM_DOCUMENTATION.md.
A Dockerfile is provided for running the service in an isolated environment with pre-configured dependencies, especially useful for ensuring consistent FFmpeg and CUDA (if configured) support across Windows and Linux hosts.