We use cookies to improve your browsing experience and to analyze our website traffic. By clicking “Accept All” you agree to our use of cookies. Privacy policy.

Story-Story: An Agentic AI-Driven Storytelling System for African Historical Narratives

Story-Story: An AI-Driven Storytelling System for African Historical Narratives

Introduction

Storytelling plays a crucial role in cultural preservation, education, and entertainment. However, many African historical figures remain underrepresented in children's literature. Secondly, in Nigeria, the curriculum does not give appropriate attention to history which is very vital. As a result, children grow uninterested in learning about their roots and national heroes.

Story-Story is an AI-powered storytelling system designed to generate engaging and factual narratives about African historical figures for young readers. The system employs an agentic workflow powered by large language models (LLMs) and diffusion-based image generation techniques to create comprehensive, child-friendly stories. The system's architecture, agent roles, and implementation details are outlined and discussions on its potential impact on education and cultural representation are made.

Our system takes in the name of a historical figure and autonomously generates a 10-page illustrated story, ensuring factual accuracy and cultural relevance. The key contributions of this work include:

An AI-driven storytelling pipeline that generates structured, coherent, and historically accurate narratives.
A multimodal approach combining text-based story generation with AI-assisted illustrations.
A content verification agent to ensure safe child-friendly storytelling.

2. System Architecture

Story-Story is built on an ensemble of AI agents orchestrated using LangGraph. The system consists of five primary agents:

2.1 Planner Agent

Role: Generates a structured outline (chapter titles) for the story.
Input: historical_figure
Output: chapter_title
Model: GPT-4o

2.2 Storyteller Agent

Role: Generates detailed, historically accurate chapter content.
Input: historical_figure, chapter_title
Output: chapter_content
Model: GPT-4o

2.3 Illustrator Agent

Role: Creates prompts for generating illustrations based on chapter content.
Input: historical_figure, chapter_title
Output: prompt
Model: GPT-4o

2.4 Image Generator

Role: Produces AI-generated illustrations from the prompts.
Input: historical_figure, prompt
Output: image
Model: Flux.1 Schnell

2.5 Publisher Agent

Role: Ensures the final story and images adhere to child-friendly content guidelines.
Input: historical_figure, chapter_content, image
Output: score
Model: GPT-4v

2.6 Workflow Overview

The entire process follows a sequential workflow:

The Planner Agent generates a structured outline for the story.
The Storyteller Agent expands each chapter title into a detailed narrative.
The Illustrator Agent formulates image generation prompts.
The Image Generator creates illustrations using diffusion models.
The Publisher Agent verifies the appropriateness of content before finalization.

3. Implementation

3.1 Technology Stack

The system is built using:

LangGraph for agent coordination
OpenAI GPT-4o & GPT-4v for story generation and verification
Flux.1 Schnell for AI-assisted illustration
Supabase for database storage
Python for backend logic

3.2 Running the System

3.2.1 Installation

pip install -r requirements.txt

3.2.2 Running the Application

python main.py --historical_figure "Ahmadu Bello"

This command generates a complete story with illustrations, storing the output in a Supabase database.

4. Database Structure

The system uses Supabase to store generated stories. The database consists of two main tables:

4.1 Stories Table

Column	Type	Description
id	Primary Key	Unique identifier for story
title	String	Title of the generated story
historical_figure	String	Name of the historical figure
description	String	Brief story description
created_at	Timestamp	Date of creation

4.2 StoryPages Table

Column	Type	Description
id	Primary Key	Unique identifier for each page
story_id	Foreign Key	References `Stories.id`
page_number	Integer	Page number in the story
chapter	Text	Chapter content
illustration_url	String	URL to the illustration
prompt	Text	AI-generated prompt for illustration

5. Results and Discussion

5.1 Generated Stories

Story-Story successfully produces engaging and informative narratives for African historical figures. The content is structured, historically accurate, and suitable for children. The generated illustrations complement the stories, enhancing engagement.

5.2 Strengths

Automated Story Generation: The system autonomously produces full-length, illustrated stories from a single input.
Child-Friendly Curation: Ensures content is appropriate and engaging.
Cultural Representation: Promotes African historical figures in children's literature.

5.3 Limitations & Future Work

While Story-Story provides a robust storytelling framework, the following enhancements are planned:

Implementing Retrieval-Augmented Generation to improve factual accuracy by integrating a bigger pool of documented history.
Adding text-to-speech with Kokoro TTS for an immersive audio experience.
Expanding to other African languages.

6. Conclusion

Story-Story showcases the power of AI-driven storytelling in bringing African historical narratives to children. Given the recent advancements in LLM agents and multimodal AI, a platform such as Story-Story is feasible and provides a fresh and engaging way to automate storytelling, enhancing education and cultural understanding.

Future enhancements, such as RAG-boosted accuracy and the addition of multiple languages, will solidify its integrity and make the delivery of African literature more seamless.