Comparative Study of Text-to-Image Models: A Focus on Subject-Specific Training for Improved Generation

• ScholarWorks: https://scholarworks.calstate.edu/concern/projects/pz50h376p
• Google Scholar: https://rb.gy/0pnc5r

Running locally

Installing the dependencies

1. Clone repo: git clone https://github.com/csulb-datascience/Fine-Tuning-on-Stable-Diffusion.git
2. Start a virtual environment of your choice. Recommended: conda
3. Install Python packages: pip install -r requirements.txt
4. Configure your training environment: accelerate config
5. Login to your HuggingFace account: huggingface-cli login
6. (optional) Login to your Weights & Biases account for training metrics: wandb login

Launching the training script

1. Navigate to training folder: cd /training
2. Choose a fine-tuning script and edit variables accordingly
3. Run: ./finetune.sh

References

• huggingface/diffusers
• Low-Rank Adaptation of Large Language Models ()
• runwayml/stable-diffusion-v1-5
• CompVis/stable-diffusion-v1-1
• lambdalabs/pokemon-blip-captions
• zoheb/sketch-scene
• hahminlew/kream-product-blip-captions