Tori is an AI system that turns a child’s diary into a personalized fairytale book — complete with story, illustrations, and background music.
Children can write daily diary entries, transform them into unique storybooks using generative AI, and share them with friends.

🔗 Live Website: https://tori-fairytale.vercel.app/
(ID: test / PW: a1234567!)
Note: The live prototype (2023) is no longer hosted; the full pipeline can be reproduced locally using this repository.

✅ Converts children’s diary entries into full fairytale stories
✅ Generates AI illustrations for every paragraph
✅ Produces background music aligned with story mood
✅ Web platform to read and share storybooks
✅ Entire system deployed on cloud GPU server

1. Child writes a diary on the web platform
2. HyperCLOVA X / ChatGPT generates the full fairytale narrative
3. Each paragraph is sent to DALL·E to generate illustrations
4. MusicGen produces a custom BGM based on story context
5. The final book is displayed on the website and shared

tori-ai/
├── app/                     # FastAPI services
│   ├── illustration_deploy.py   # DALL·E illustration generation
│   ├── hyperclova_deploy.py     # Story generation (HyperCLOVA X)
│   └── music_deploy.py          # MusicGen BGM generation
├── nginx/                   # Reverse proxy configuration
└── textual_inversion_project/   # SD experiments (LoRA, Textual Inversion)

• Ubuntu 20.04 LTS (64-bit)
• GPU: Tesla T4
• Nvidia Driver 535.129.03
• CUDA 12.2
• Python 3.8
• FastAPI + Gunicorn + Nginx
• S3 bucket for storing generated images / BGM

git clone https://github.com/6garlics/tori-ai.git
cd tori-ai
pip install -r requirements.txt

sudo apt update && sudo apt upgrade
sudo apt autoremove
sudo apt install nginx
sudo cp ${PROJECT_ROOT}/nginx/nginx.conf /etc/nginx/nginx.conf

sudo nginx -t
sudo systemctl start nginx

Create secrets.json in ${PROJECT_ROOT}/app with OpenAI, HyperCLOVA, S3 keys, endpoints, etc.

# Illustration service
python app/illustration_deploy.py

# Story generation service
nohup python app/hyperclova_deploy.py &

# Music generation service
nohup python app/music_deploy.py &

kill -9 $(lsof -i:${PORT} -t) 2>/dev/null

• Provisioned on Tencent Cloud
• Nginx reverse proxy with HTTPS (SSL enabled)
• Gunicorn + FastAPI backend
• Integrated S3 for media storage
• Connected OpenAI API (ChatGPT, DALL·E)
• Integrated HyperCLOVA X API for Korean storytelling

• Story generation from children’s diaries
• Illustration generation using DALL·E
• BGM generation with MusicGen
• Fully automated workflow from diary → illustrated fairytale → sharable book

• Fine-tuned Stable Diffusion using Textual Inversion
• Fine-tuned Stable Diffusion using LoRA
• (Later deprecated in production in favor of DALL·E)

• Style-consistent character generation across pages
• Inference speed optimization on GPU
• Add multilingual diary → storybook support
• Real-time audiobook narration