Project information

• Category: Model Training

This project focuses on training a LoRA model for Ron Perlman's Hellboy. I chose this character because it's very complex, with asymmetrical arms, a trenchcoat, utility belt, and numerous elements to capture properly. LoRA, or Low-Rank Adaptation, enables efficient fine-tuning of diffusion models for specific concepts, like characters, with smaller file sizes and faster training times.

I gathered a diverse dataset of Ron Perlman's Hellboy, curated for quality and relevance. I also had to restore some images where the face looked blurry in the distance to ensure the model works well in general shots. Captions describing pose, expression, and background were created for each image, serving as prompts for the LoRA model. Training was done on Stable Diffusion XL, optimizing settings and hyperparameters for quality.