diffusion model trained on my art

I trained a Flux LoRA on 20 samples of my artwork spanning oil, acrylic, watercolor, ink, charcoal, and colored pencil. I built a custom labeling pipeline using GPT-4o to generate training captions, a Flask backend for inference via Fal AI, and a web interface for real-time generation. The project investigates which elements of my artistic style transfer most effectively through fine-tuning.

The artwork depicts a bustling kitchen scene with three individuals engaged in various activities around the room. The composition centers around the kitchen's interior, filled with detailed cabinetry, appliances, and everyday objects spread across the countertops. The individuals, portrayed in mid-action, contribute to the dynamic movement throughout the piece. Rendered in graphite, the monochromatic color palette emphasizes depth through shading and highlights, creating a strong contrast between light streaming through a window and the shadows within the room. The lighting is natural, pouring from the window, casting soft, elongated shadows across the space, enhancing the realism. The medium utilized is charcoal, noted for its ability to produce rich, dark tones and soft transitions, contributing to an intimate atmosphere. The mood conveys domestic tranquility and routine, with a sense of warmth and familiarity inherent in the setting. Notable artistic techniques include cross-hatching and smudging, providing texture and form to the figures and objects.