🎐 Japanese Haiku Generator App

This is a student project that was created for traditional Japanese haiku poems generation in response to user-defined themes — with translations and authentic Ukiyo-e style illustrations. The main aim was to wrap up some knowledge obtained during ‘LangChain for LLM Application Development’ AI Course and practice project architecture building.

---

🎯 The Model Choice

It was decided to pick a cohere LLM model for it has an intuitively clear and coherent API. Moreover, the model demonstrates great results in creativity and handles both Russian and Japanese text generation really well.

---

🌸 Features

• 🖋️ Haiku Generator (in Japanese) using Cohere language models
• 📜 Automatic Russian Translation of generated haiku
• 🖼️ Ukiyo-e Style Image Generation via Stable-Diffusion 3.5 large
• 🧠 Theme validation with simple heuristics
• ✨ Minimalistic Streamlit UI
• 📁 Modular, extensible project structure

---

🌸 Example Output

Input Theme: Танабата

Generated Haiku (日本語):

星々交わる
夏の夜にたなばたの
夢描く人

Russian Translation:

Звёзды переплетаются
В летней ночи, Танабата
Люди рисуют мечты.

Generated Image:

---

💻 Technologies Used

Tech	Description
🐍 Python 3.10+	Core language
🌐 Streamlit	Frontend web app
🧠 Cohere	Haiku generation & translation
🎨 Hugging Face API	stable-diffusion 3.5 large image generation
📦 dotenv	Local API token management

---

🛠 Installation

git clone https://github.com/yourusername/haiku-generator.git
cd haiku-generator
python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows
pip install -r requirements.txt

---

🔐 Environment Variables

Create an .env file in the root directory containing the paramount API keys:

COHERE_API_KEY='your_cohere_key'
HF_TOKEN='your_huggingface_token'

---

🚀 Running the App

streamlit run app.py

---

📁 Project Structure

haiku-generator/
├── app.py -> Streamlit app entry point
├── haiku_generator.py -> Japanese haiku generation (Cohere)
├── russian_translation.py -> Poetic Russian translation
├── ukiyo_e_generator.py -> Image generation (stable-diffusion 3.5 large)
├── theme_validation.py -> Theme quality checker
├── requirements.txt
└── .env

---

❗ Notes

A Cohere API is geo-blocked in certain regions, like Russia. So if its implementation is unavailable in your region consider using a foreign server or a VPN software.