Content Creation Automation Pipeline

⚠️ DISCLAIMER: This project is a working prototype ("brouillon") and is highly specific to a particular environment and workflow. It was developed for personal use and is shared here for inspiration, reference, or adaptation. Expect hardcoded paths, specific folder structures, and a "raw" coding style.

📌 Overview

This project is an end-to-end automated pipeline designed to create and publish short-form video content (TikToks, Instagram Reels, YouTube Shorts). It leverages AI for scriptwriting and metadata, Remotion for programmatic video editing, and Telegram for final delivery.

🏗️ Architecture & Workflow

The pipeline is orchestrated by a main shell script (automate.sh) that connects several specialized modules:

1. Script Generation (creator/): Uses the Google Gemini API to generate scripts based on specific "angles" (e.g., Stoic, Cynical, Wise) and themes.
2. Asset Selection (selector/): Python scripts that select appropriate video backgrounds and assets based on the content type.
3. Video Assembly (assembler/): A Remotion project that renders the base video by sequencing selected clips and adding background audio.
4. Transcription & Subtitles (captioner/):
   • Uses Whisper (via Node.js) to transcribe the generated audio.
   • A second Remotion project applies styled, animated subtitles to the video.
5. Metadata & Thumbnail (deliver/):
   • AI Metadata: Gemini extracts the best hook for the description and selects relevant hashtags.
   • Dynamic Thumbnails: Python (Pillow) generates a custom thumbnail image with high-CTR text overlays.
6. Delivery (deliver/): Automatically sends the final MP4, description, and thumbnail to a designated Telegram chat via a Bot.

🚀 Key Features

• Multi-Character Templates: Supports different visual/audio styles like "Default", "Oogway" (Wise), or "Joker" (Dark/Cynical).
• AI-Powered Copywriting: Automatic generation of hooks, descriptions, and hashtags.
• Dynamic Video Rendering: Programmatic editing via Remotion (React-based).
• Automated Triggers: Includes scripts for daily automated runs with rotating themes.
• Telegram Integration: Immediate notification and file delivery upon completion.

🛠️ Prerequisites

To run this pipeline, you will need:

• Node.js & npm (for Remotion and Whisper scripts)
• Python 3.10+ (for AI logic and image processing)
  • see requirements.txt
• FFmpeg (essential for Remotion and audio processing)
• yt-dlp (for downloading audio assets)
• API Keys:
  • Google Gemini API Key
  • Telegram Bot Token

⚙️ Setup & Configuration

1. Environment Variables: Create a .env file in the root directory. Use the provided code structure as a reference for required variables:

   • GEMINI_API_KEY: Your Google AI Studio key.
   • TOKEN: Your Telegram Bot token.
   • URL: Your Telegram Bot API URL (or standard API).
   • PYTHON_PATH, YT_DLP_PATH, etc.: Paths to your local executables.
   • PROJECT_BASE_PATH: The absolute path to the project root.
   • Cf .env.bak for other needed variables

2. Dependencies:

   # Install Python deps
   pip install google-genai requests python-dotenv Pillow
   
   # Install Node.js deps in assembler and captioner folders
   cd assembler && npm install
   cd ../captioner && npm install
   

📂 Project Structure

• automate.sh: The main entry point script.
• assembler/: Remotion source code for the raw video composition.
• captioner/: Remotion source code for subtitling logic.
• creator/: Python logic for Gemini-powered scriptwriting.
• deliver/: Python scripts for thumbnail generation and Telegram uploading.
• selector/: Logic for picking video background clips.
• trigger/: Daily automation triggers and theme pools.

⚠️ Important Notes

• Hardcoded Paths: Many scripts rely on environment variables pointing to specific local directories. Ensure your .env is perfectly configured.
• Specific Assets: The project expects certain font files (e.g., Lora-SemiBold.ttf) and template images to exist in the deliver/ and captioner/public/ folders.
• Customization: You will likely need to modify the system_instruction in script-creator.py and the prompts in deliver.py to match your specific niche.
• Local ressources: this pipeline relies on several local ressources such as videos sources, audio etc... You can see an example of the local tree files in the tree.txt file.