Mitchins/t5-base-artgen-bidirectional

T5-Base AI Art Prompt Generator

A fine-tuned T5-base model for bidirectional AI art prompt transformation, trained on 48K+ high-quality prompts with advanced bias protection.

🚀 Quick Start

# Test the model interactively
python3 ../test_model.py --model fine_tuned_t5_base --interactive

# Run batch tests
python3 ../test_model.py --model fine_tuned_t5_base --batch

# Single transformations
python3 ../test_model.py --model fine_tuned_t5_base --elaborate "cat sitting"
python3 ../test_model.py --model fine_tuned_t5_base --simplify "detailed cat prompt..."

📊 Model Stats

• Parameters: 220M (T5-base)
• Training Samples: 48,034
• Validation Loss: 0.4293
• Training Epochs: 5
• Bias Protection: ✅ Active

🎯 Capabilities

• Simple → Elaborate: Transform basic descriptions into detailed art prompts
• Elaborate → Simple: Extract core concepts from complex prompts
• Multi-Platform: Trained on NightCafe, Civitai, and community data
• Quality Filtered: Advanced filtering for high-quality outputs

📁 Files

• MODEL_CARD.md - Comprehensive model documentation
• config.json - Model architecture configuration
• model.safetensors - Model weights (850MB)
• training_info.txt - Training metrics and parameters
• *.json - Tokenizer and generation configurations

🔧 Requirements

• Python 3.8+
• Transformers 4.20.0+
• PyTorch 1.10.0+
• 2GB+ RAM (4GB GPU recommended)

📈 Performance

Compared to T5-small:

• Quality: Significantly improved detail and coherence
• Speed: ~2.3x slower inference
• Memory: ~850MB GPU memory usage
• Bias: Better diversity and reduced oversaturation

See MODEL_CARD.md for detailed performance analysis and examples.