Textual Inversion for Personalization of Diffusion-Based Image Generation.

This project explores the personalization of the Stable Diffusion model for text-to-image generation through textual inversion, focusing on African artistic styles. We fine-tuned the model using a dataset comprising LAION 5B and images from three African artists, resulting in the generation of culturally specific images. The model’s performance was evaluated using Mean Opinion Scores, revealing above-average success in incorporating the targeted artistic styles. The findings highlight the model’s robustness in generating text-conditioned images that reflect diverse and specific cultural influences, with potential for further enhancement in style representation and training efficiency.