Skip to content

An automated educational video generation pipeline powered by Gemini 3. It transforms any topic string into a full sketch based animated video using a complex multi-agent workflow.

License

Notifications You must be signed in to change notification settings

sayan404/vibe_video

Repository files navigation

VV Pipeline (Research → Storyboard → Manim → Render → Critique → Iterate → Export)

An automated educational video generation pipeline powered by Gemini. It transforms a topic string into a full Manim animation using a multi-agent workflow.

Features

  • Research Agent: Deconstructs topics into learning objectives.
  • Storyboard Agent: Generates visual plans (frames, intent, voiceover).
  • Coding Agent: Writes Manim (Python) code.
  • Render Engine: Compiles code to video.
  • Critique & Repair Loop:
    • Critique Agent: Watches the video to find visual/timing issues.
    • Repair Agent: Automatically fixes syntax/runtime errors during rendering.
    • Iteration Agent: Refines the animation code based on critique feedback.

Prerequisites

  1. Node.js (v18+)
  2. Python (3.10+)
  3. FFmpeg (Required by Manim)
  4. LaTeX (Optional, but recommended for Manim text rendering)

Setup

1. Install Dependencies

Node.js:

npm install

Python: Create a virtual environment (recommended) and install Manim:

# Windows
python -m venv venv
.\venv\Scripts\activate
pip install manim google-genai

# Mac/Linux
python3 -m venv venv
source venv/bin/activate
pip install manim google-genai

2. Configure Environment

Copy the example file and add your keys:

cp .env.example .env

Edit .env:

GEMINI_API_KEY=your_api_key_here
# Optional: Override defaults
# GEMINI_MODEL=gemini-2.0-flash-exp
# PYTHON=python 

Note: Ensure the PYTHON var in .env points to your virtualenv python if unrelated to your system path.

Running the App

Start the Dashboard

npm run dev:web

Open http://localhost:3000.

Generating a Video

  1. Go to the dashboard.
  2. Click "New Run".
  3. Enter a topic (e.g., "Binary Search", "Photosynthesis").
  4. The pipeline will execute phases sequentially:
    • Phase 1: Research
    • Phase 2: Storyboard
    • Phase 3: CodeGen
    • Phase 4: Render (Self-Healing active)
    • Phase 5: Critique
    • Phase 6: Iteration (if needed)
    • Phase 7: Export

Project Structure

  • apps/web: Next.js frontend.
  • services/: Core logic for each phase.
    • gemini/: Research & CodeGen agents.
    • nanobanana/: Storyboard agent.
    • critique/: Video analysis agent.
    • iteration/: Refinement & Repair agents.
    • pipeline/: Orchestratation logic (export.ts).
  • python/: Manim renderer entrypoint.
  • runs/: Output artifacts for every run.

Troubleshooting

  • Manim Render Error: The pipeline has a self-healing loop (Phase 4). Check runs/<id>/phase4.repair.log or the dashboard status.
  • FFmpeg not found: Ensure ffmpeg is in your system PATH.

About

An automated educational video generation pipeline powered by Gemini 3. It transforms any topic string into a full sketch based animated video using a complex multi-agent workflow.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published