I Built an AI Video Factory That Runs 24/7 — Fully Open Source

arya rizky — Sat, 09 May 2026 08:50:05 +0000

I wanted to share a project I've been building: Mesin Cuan Viral Architect — a fully automated YouTube content pipeline.

What it does:
Discovers trending topics, writes scripts with dual parallel AI, generates TTS voiceover, fetches B-roll footage, renders cinematic video via FFmpeg, and uploads to YouTube — all autonomous, 24/7.

The secret sauce — Dual Parallel AI Scripting:
Most AI content tools use a single LLM → generic output. Mesin Cuan runs Qwen + Ollama simultaneously, then they cross-score each other to eliminate AI slop before production. The higher-scoring script wins.

12-engine pipeline:

Viral Loop Engine — real-time trend detection
Dual Parallel AI — Qwen + Ollama cross-scoring
Script Quality Scorer — multi-dimensional validation
Hook Engine — auto-injects high-retention openers
Memory Engine — never repeats content
Research Engine — web research before scripting
Series Engine — auto-generates multi-part content
Neon Visuals — cinematic FFmpeg rendering
Smart SFX Mixer — niche-aware sound effects
Thumbnail Intelligence — AI-driven CTR optimization
OAuth2 Analytics — per-channel retention dashboard
Pipeline Estimator — ETA prediction for batch renders

Why I built this:
I was spending 8+ hours editing a single video. Hiring editors was too expensive and existing AI tools produced generic content. So I built my own video factory.

Tech stack: Python 3.11 + FFmpeg + Ollama + Qwen + F5-TTS

Quick start (only needs Ollama + FFmpeg):

git clone https://github.com/algojogacor/mesin-cuan.git
cd mesin-cuan
python main.py --channel ch_id_horror --skip-qc

Repo: https://github.com/algojogacor/mesin-cuan

MIT License — free to use, modify, distribute. Built solo at 18.

DEV Community: arya rizky

I Built an AI Video Factory That Runs 24/7 — Fully Open Source