KODI: A Korean Diffusion Model for Bilingual Text-to-Image Generation and Cultural Fidelity

KODI is a diffusion model that generates high-quality Korean cultural images from Korean text prompts. It uses a Korean CLIP-based text encoder to better understand Korean prompts and generate culturally appropriate images.

📦 Installation

git clone https://github.com/TeamLab/kodi.git
cd kodi
pip install -r requirements.txt

📊 Datasets and Benchmarks

Korean Cultural Dataset (KCD)

Our Korean cultural training dataset is located at:

korean-cultural-dataset/

This dataset contains Korean cultural images with corresponding Korean text descriptions.

MC-K Evaluation Benchmark

The Korean cultural evaluation benchmark MC-K is available at:

evaluation/benchmark/

This benchmark is used for evaluating cultural appropriateness and Korean language understanding.

🔧 Usage

1. Model Training

# Train KODI with Korean Cultural Dataset (KCD)
python training/train_kodi.py --config training/configs/kodi.py

2. Image Generation

# Generate images with evaluation dataset
python evaluation/generate_eval_images.py

3. Model Evaluation

# Evaluate with MC-K benchmark using KC-CLIP
python evaluation/evaluate_by_kcclip.py

🤗 Model Weights

Model	Type	Location	Description
KODI	LoRA Weights	model-weights/kodi/	Korean cultural diffusion model (included in repository)
KC-CLIP KO	Evaluator	letgoofthepizza/kc-clip-ko	Korean cultural CLIP model (Korean)
KC-CLIP EN	Evaluator	letgoofthepizza/kc-clip-en	Korean cultural CLIP model (English)