🎬 Script2Reel

A Provider-Agnostic Multi-Agent AI Framework for Short-Form Video Generation

Transform a single prompt into a fully narrated short-form video using specialized AI agents, interchangeable AI providers, and an extensible workflow.

---

Built with

LangGraph • Streamlit • Python • Gemini • OpenAI • Stable Diffusion XL • ElevenLabs • Edge-TTS

---

Overview

Script2Reel is an AI-powered content generation framework that automatically converts a single text prompt into a narrated short-form video suitable for Instagram Reels, YouTube Shorts, or TikTok.

Unlike traditional text-to-video systems that rely on one large prompt, Script2Reel decomposes the problem into multiple specialized AI agents.

Each agent has a single responsibility:

• understand the prompt
• write the screenplay
• plan scenes
• review consistency
• generate visual prompts
• generate images
• generate narration
• assemble the final video

This modular workflow makes the system easier to improve, debug and extend.

---

Features

Multi-Agent Workflow

• Intent Classification
• Creative Director
• Educational Writer
• Screenwriter
• Scene Planner
• Consistency Reviewer
• Subject Extractor
• Visual Director
• Image Generation
• Voice Generation
• Video Assembly

---

Provider Agnostic

Supports interchangeable AI providers.

Current implementations:

Category	Providers
LLM	Gemini, OpenAI
Image	Local SDXL, OpenAI, Mock
Voice	ElevenLabs, Edge-TTS
Video	Local, Mock

The workflow is completely independent of the underlying provider.

---

Intelligent Infrastructure

• Provider Registry
• Automatic Provider Discovery
• Automatic Provider Selection
• Automatic Fallback
• Provider Capability Metadata
• Startup Health Checks
• Runtime Telemetry
• Benchmark Recommendations

---

Workflow

User Prompt
      │
      ▼
Intent Classification
      │
      ▼
Creative / Educational Route
      │
      ▼
Screenwriter
      │
      ▼
Scene Planner
      │
      ▼
Consistency Review
      │
      ▼
Subject Extraction
      │
      ▼
Visual Director
      │
      ▼
Image Generation
      │
      ▼
Voice Generation
      │
      ▼
Video Assembly
      │
      ▼
Final MP4

---

Architecture

                 LangGraph Workflow
                         │
                         ▼
              AI Agent Pipeline
                         │
                         ▼
               Provider Abstraction
        ┌────────┬────────┬────────┬────────┐
        │  LLM   │ Image  │ Voice  │ Video  │
        └────────┴────────┴────────┴────────┘
                         │
                         ▼
            Local or Cloud AI Providers

---

Why this architecture?

Instead of tightly coupling the workflow to one model provider, Script2Reel separates:

• workflow orchestration
• AI providers
• rendering
• configuration

This allows providers to be swapped without modifying the workflow itself.

---

Screenshots

🚀 Landing Page

The application starts with a simple interface where users can create AI-generated reels from a single prompt.

✍️ Reel Configuration

Users can configure the prompt, duration, style, platform, voice, and generation mode.

🤖 Multi-Agent Workflow

During generation, every AI agent executes independently.

The progress tracker exposes each stage of the pipeline.

🎬 Final Reel

The generated reel can be previewed, downloaded, and inspected.

---

Quick Start

Clone

git clone https://github.com/AkibDa/Script2Reel.git

cd Script2Reel

Install

python -m venv .venv

source .venv/bin/activate

pip install -r requirements.txt

---

Environment Variables

GOOGLE_API_KEY=

OPENAI_API_KEY=

ELEVENLABS_API_KEY=

FLUX_API_KEY=

OLLAMA_HOST=

Only configure the providers you intend to use.

---

Run

uvicorn backend.main:app --host 127.0.0.1 --port 8000

---

Documentation

• 📖 ARCHITECTURE.md
• 🤝 CONTRIBUTING.md

---

Roadmap

• Image-to-video generation
• Temporal scene animation
• Kokoro TTS
• FLUX support
• Wan2.1 support
• Veo support
• Additional provider implementations
• Scene editing interface

---

License

MIT License