youtube-agentic-ai-studio README

The project's README from github.com/raunakpatil/youtube-agentic-ai-studio. As one markdown file: README.md.

YouTube AI Agent Studio

🎬 Ai Powered Free Automated Faceless Youtube Studio

A fully automated, 100% free AI pipeline that researches, scripts, narrates, animates, and uploads faceless YouTube videos — from a single command.

Python Gemini License: MIT Free Tier Stars

Quick Start · Features · Pipeline · Setup Guide · FAQ

✨ Features

Feature Details
🧠 AI Research Gemini scans for trending topics tailored to your channel niche
✍️ Script Writing Full narrated script with hook, sections, and CTA — in JSON
🎙️ Free Voice Microsoft Edge TTS (no API key, no cost, word-level timing)
🖼️ Stock Images Auto-downloads from Pexels with topic-aware search
🎞️ Animated Video Ken Burns zoom, smooth crossfades, captions, progress bars
📱 Shorts Support Generates vertical 9:16 Shorts alongside the main video
🕵️ Human Review Local web dashboard to watch and approve before publishing
🚀 Auto Upload Publishes to YouTube with SEO tags, description, and thumbnail
🔄 Smart Fallback 8-model chain auto-switches if any model hits quota
🎵 Background Music Drop MP3s in music library/ — auto-looped to fit video length

🖥️ Screenshots
Main GUI Settings
Main dashboard — generate videos with one click Settings panel — configure every aspect
Review Dashboard Banner
Review dashboard — watch before you publish Pipeline overview

🎬 Live Example — Built With This Tool

This repo powers Eldritch Lore — a real YouTube channel running entirely on this pipeline.

YouTube Subscribers

"What if Napoleon had nukes? What if Earth was flat... and spinning?" Every video on this channel was researched, scripted, narrated, and uploaded using exactly this code.

→ See what this tool can actually produce

⚡ Quick Start

# 1. Clone
git clone https://github.com/raunakpatil/youtube-agentic-ai-studio.git
cd youtube-agentic-ai-studio

# 2. Install dependencies
pip install -r requirements.txt

# 3. Set your API keys (copy and edit .env.example)
cp .env.example .env
# Edit .env and add your GEMINI_API_KEY and PEXELS_API_KEY

# 4. Run the GUI
python gui.py
# → Opens at http://localhost:7842

# — OR — run the pipeline directly from the terminal
python pipeline.py

First run? Read the full Setup Guide — it takes about 10 minutes.

🏗️ How It Works

The pipeline runs 6 sequential phases:

python pipeline.py  ──►  Research  ──►  Script  ──►  Narration
                                                          │
              YouTube ◄── Upload ◄── Review ◄── Video ◄──┘
Phase Agent What it does
1 · Research agents/researcher.py Gemini brainstorms trending topics for your niche and outputs a structured brief
2 · Script agents/scriptwriter.py Expands the brief into a full narrated script with image queries per section
3 · Narration video/narrator.py Edge-TTS converts the script to a cinematic MP3 voiceover
4 · Images video/stock.py Downloads high-res stock photos from Pexels, topic-aware per section
5 · Video video/creator.py Renders animated slides with Ken Burns, crossfades, captions, and music
6 · Review review/app.py Spins up a local dashboard — you watch and approve before anything is published
7 · Upload uploader/youtube.py Publishes to YouTube with full SEO metadata on your approval

📁 Project Structure

youtube-ai-agent/
├── 📄 pipeline.py              ← Run this to start the full pipeline
├── 🖥️  gui.py                  ← Run this for the visual GUI
├── ⚙️  config.py               ← All settings: keys, colors, voice, video
├── 📦  requirements.txt
├── 🔑  client_secret.json      ← YouTube OAuth (you add yours — see SETUP.md)
├── 📝  .env.example            ← Copy to .env and fill in your API keys
│
├── agents/
│   ├── gemini_client.py        ← Shared AI client with 8-model fallback chain
│   ├── researcher.py           ← Topic research agent
│   └── scriptwriter.py        ← Script writing agent
│
├── video/
│   ├── narrator.py             ← Edge-TTS voiceover generator
│   ├── stock.py                ← Pexels image downloader
│   ├── creator.py              ← Full video renderer (Ken Burns, captions, etc.)
│   └── music.py                ← Background music mixer
│
├── review/
│   ├── app.py                  ← Flask review server
│   └── templates/review.html  ← Review dashboard UI
│
├── uploader/
│   └── youtube.py              ← YouTube Data API v3 upload
│
├── gui_templates/
│   └── index.html              ← Full GUI (served by gui.py)
│
├── music library/              ← Drop your MP3/WAV tracks here
│
└── output/                     ← Auto-generated (gitignored)
    ├── research.json
    ├── script.json
    ├── narration.mp3
    ├── images/
    └── final_video.mp4

🤖 AI Model Fallback Chain

The system automatically tries models in order, rotating from your Starting Model preference:

gemini-2.5-flash  →  gemini-2.5-flash-lite  →  gemma-4-31b-it
       ↓
gemini-2.0-flash  →  gemini-2.0-flash-lite  →  gemma-3-27b-it
       ↓
gemini-1.5-flash-001  →  gemini-1.5-pro-001

If a model hits quota (429) or is overloaded (503), it automatically retries up to 3 times then advances to the next model — so your pipeline never stops mid-run.

Set your preferred starting model in the GUI under Settings → AI Model, or in config.py:

GEMINI_MODEL = "gemini-2.0-flash"   # 1500 free req/day — recommended default

🔑 API Keys Required (All Free)

Service Key Free Tier Where to get it
Google Gemini GEMINI_API_KEY 1500 req/day aistudio.google.com
Pexels PEXELS_API_KEY 200 req/hour pexels.com/api
Edge TTS (none) Unlimited Built-in
YouTube API client_secret.json Free Google Cloud Console
ElevenLabs ELEVENLABS_API_KEY Optional elevenlabs.io

Total cost: $0 per video. All required services have free tiers that comfortably cover daily video production.

🎨 Customising Your Channel

Edit config.py to make the output yours:

# 1. Describe your channel — the more specific, the better the topics
CHANNEL_DESCRIPTION = """
An educational channel covering the psychology of decision-making.
Target audience: adults 25-40 who enjoy Lex Fridman and Hidden Brain.
"""
CHANNEL_NAME = "Mind Mechanics"

# 2. Pick your voice
VOICE_ID = "en-GB-RyanNeural"   # British, deep, cinematic

# 3. Brand colours (RGB tuples)
COLORS = {
    "primary":   (99, 102, 241),   # your brand colour
    "highlight": (251, 191, 36),   # accent / titles
    ...
}

# 4. Video style
CROSSFADE_DURATION = 0.7   # image transition smoothness
OVERLAY_OPACITY    = 0.62  # how dark the image overlay is

❓ FAQ

Does this cost anything to run?

No. The entire stack uses free tiers:

  • Gemini — 1500 req/day free on the 2.0-flash model
  • Pexels — 200 req/hour free
  • Edge TTS — completely free, no account needed
  • YouTube Data API — free for uploads
  • Video rendering — runs locally on your machine (CPU/GPU)
How long does one video take to generate?

On a mid-range machine (i7 + 16GB RAM):

  • Research + Script: ~30 seconds
  • Narration: ~10 seconds
  • Image downloads: ~20 seconds
  • Video render (10 min video): ~3–8 minutes depending on RENDER_PRESET
  • Upload: ~2–5 minutes depending on internet speed
Can I use this without the YouTube upload?

Yes. The review step lets you click "Reject" to skip the upload — the rendered MP4 will still be saved in output/. You can also comment out the upload step in pipeline.py entirely.

Why is moviepy pinned to 1.0.3?

MoviePy v2.0 removed the moviepy.editor module and changed the entire API. v1.0.3 is required. Similarly, NumPy must be <2.0.0 because MoviePy's underlying C extensions were compiled against NumPy 1.x. Do not upgrade either.

The video renders with tiny/ugly text

Your system is missing the fonts the renderer looks for. Install DejaVu fonts:

  • Ubuntu/Debian: sudo apt install fonts-dejavu
  • Mac: brew install --cask font-dejavu-sans-mono-nerd-font
  • Windows: fonts are already present in C:/Windows/Fonts/
YouTube upload fails with "access denied"

Delete youtube_token.pickle and re-run. This forces a fresh OAuth flow. Make sure your Google account is added as a Test User in Google Cloud Console → OAuth consent screen.

🗺️ Roadmap

🤝 Contributing

Pull requests are welcome! Please open an issue first to discuss major changes.

  1. Fork the repo
  2. Create a feature branch: git checkout -b feature/my-feature
  3. Commit: git commit -m 'Add my feature'
  4. Push: git push origin feature/my-feature
  5. Open a Pull Request

🌍 Made With This Tool

Eldritch Lore — science, history, and "what if" scenarios. Running 100% on this pipeline. Subscribe to see it in action.

Made with ❤️ by @raunakpatil

Star on GitHub

youtube-agentic-ai-studio (raunakpatil/youtube-agentic-ai-studio)

Searching Database…

137 ·GitHub
Youtube-agentic-ai-studio
Что это?Полностью автоматический бесплатный AI-пайплайн, который исследует трендовые темы, пишет сценарий с дикторским текстом, генерирует озвучку и анимированное видео со стоковыми изображениями и загружает готовое видео на YouTube — всё запускается одной командой.
Это консольный инструмент или у него есть GUI?Запускается одной командой, но в комплекте есть локальный веб-дашборд с панелью настроек для конфигурации канала и экраном ревью, где можно смотреть и одобрять видео перед публикацией.
Как это запустить?Склонируйте репозиторий и пройдите Setup Guide, на который уходит около 10 минут: установите Python-зависимости, впишите API-ключи в конфиг, затем запустите команду пайплайна.
Какие API-ключи мне нужны?Нужен бесплатный ключ Google Gemini для исследований/сценариев (1500 запросов/день) и бесплатный ключ Pexels для стоковых изображений (200 запросов/час), плюс учётные данные YouTube Data API для публикации. Озвучке Edge TTS ключ не нужен, а ElevenLabs опционален.
Есть ли подводные камни при установке, о которых стоит знать?moviepy нужно держать зафиксированным на v1.0.3, потому что v2.0 удалил модуль moviepy.editor и изменил весь API; также нужны установленные шрифты DejaVu, иначе титры рендерятся крошечными и уродливыми.
Можно ли пользоваться без реальной публикации на YouTube?Да, в дашборде ревью можно нажать 'Reject', чтобы пропустить загрузку, — отрендеренный MP4 всё равно сохранится в output/, либо можно целиком закомментировать шаг загрузки в pipeline.py.
Что будет, если AI-модель упрётся в квоту посреди прогона?Пайплайн перебирает цепочку фолбэков из 8 моделей: если модель возвращает ошибку квоты или перегрузки, он делает до 3 повторов, затем автоматически переходит к следующей модели, так что прогон никогда не застревает.
Сколько времени уходит на одно видео?На средней машине (i7, 16GB RAM): исследование и написание сценария занимают около 30 секунд, озвучка ~10 секунд, скачивание изображений ~20 секунд, а рендер 10-минутного видео — примерно 3-8 минут, и загрузка добавляет ещё 2-5 минут.
Есть ли реальный канал, построенный на этом инструменте?Да — 'Eldritch Lore', YouTube-канал об истории и сценариях 'что если', целиком работает на этом пайплайне: каждое видео исследуется, пишется, озвучивается и загружается именно этим кодом.
What are similar projects?youtube-automation-agent, short-video-maker, mortiflix-oss, MoneyPrinterTurbo, OpenReels, autoclip, brag, Concat
Which topics is it in?youtube, tts, content-creation, text-to-speech, oauth, video-generation, orchestration, video, gemini, multi-agent, dashboard, free
Creator
raunakpatil
Size
8.7 MB
Stars
137
Updated
CommitOct 16
Age
5mo
Commits · 90 Days