FreakyBits — AI Video Automation Pipeline

Fully automated short-form video pipeline. Generates and uploads 6 AI-powered videos a day to YouTube Shorts and Instagram Reels. No manual work after setup.

---

How it works

GitHub Actions triggers at 7AM and 7PM IST every day. Each run produces 3 videos:

GitHub Actions (cron: 7AM + 7PM IST)
    │
    ▼
pipeline.py
    │
    ├── Gemini 2.5 Flash → viral script + SEO title + hashtags
    ├── Edge TTS (Microsoft) → neural voiceover
    ├── Pexels API → HD portrait stock clips
    │
    ▼
FFmpeg
    ├── 9:16 vertical crop
    ├── dark cinematic filter
    ├── bold white uppercase subtitles (word-by-word)
    └── muted background music mix
    │
    ▼
Upload
    ├── YouTube Shorts (YouTube Data API v3)
    └── Instagram Reels (Meta Graph API)

---

Features

What	Details
Script generation	Gemini 2.5 Flash with viral hooks, SEO titles, hashtags
Voice	Microsoft Edge TTS — Christopher (EN), Madhur (HI)
Video	Pexels HD stock API, 1080x1920, 9:16 portrait
Subtitles	FFmpeg burns bold uppercase word-by-word captions
Niches	Horror, Comedy, AI & Tech, Storytelling
Series format	Part 1 / Part 2 cliffhanger per niche for watch-through
Bilingual	EN and HI rotate across runs
Deduplication	Tracks last 500 topics — never repeats a script
Retry logic	Exponential backoff on upload failures
Analytics	Every upload logged to JSON
Cost	Runs on GitHub Actions free tier (~1,200 mins/month)

---

Setup

What you need

• GitHub account
• Google AI Studio API key (free) — aistudio.google.com
• Pexels API key (free) — pexels.com/api
• YouTube OAuth credentials via Google Cloud Console

Steps

1. Fork this repo
2. Add these GitHub Secrets (Settings → Secrets → Actions):

Secret	What it is
GEMINI_API_KEY	Google AI Studio key
PEXELS_API_KEY	Pexels API key
YOUTUBE_CLIENT_SECRETS	OAuth JSON from Google Cloud
YOUTUBE_TOKEN_B64	Base64-encoded pickle token
INSTAGRAM_TOKEN	Meta Graph API token (optional)
INSTAGRAM_ACCOUNT_ID	Instagram account ID (optional)

3. Enable GitHub Actions. Pipeline runs automatically.

---

Running tests

pip install pytest
pytest tests/ -v

---

Output

• 6 videos per day (3 per run, 2 runs)
• ~32 seconds per video
• 1080x1920px, 9:16 vertical
• 4 English + 2 Hindi per day

---

Tech stack

Layer	Tool
Script AI	Google Gemini 2.5 Flash
Voice	Microsoft Edge TTS (neural)
Video	Pexels HD Stock API
Assembly	FFmpeg
Upload	YouTube Data API v3, Instagram Graph API
CI/CD	GitHub Actions
Auth	Google OAuth 2.0

---

Project structure

freakybits/
├── pipeline.py                   # main automation pipeline
├── requirements.txt
├── tests/
│   └── test_pipeline.py
├── buzzBits_output/              # generated videos (gitignored)
│   ├── analytics.json            # upload history
│   └── used_topics.json          # deduplication store
└── .github/
    └── workflows/
        └── buzzBits_daily.yml    # cron schedule

---

Roadmap

• Oracle Cloud VM migration (no GitHub Actions minute cap)
• Gemini Vision clip relevance scoring
• Google Sheets analytics dashboard
• Instagram token auto-refresh
• Per-niche topic model fine-tuning

---

Built by

Mohammad Sahil Vahora — ECE Final Year, Parul University 2026

GitHub: @imsv1301 | Instagram: @freaky_bits