When using Generative AI such as Midjourney or Stable Diffusion, it's quite difficult to generate Khmer robam or dance images because data on Cambodian culture is limited for AI to scrape from. Through Stable Diffusion and Dreambooth, I've been training my own "diffusion model" using my own photographs and artwork.
In its infant stages, there are a lot of inadequacies and inaccuracies in the model; but it's an ambitious project; about 5% complete, I'll need to train hand gestures, poses, costumes and accessory classes, helmets and headdresses, character classes, and archetypes to build a beautiful training model that everyone could use, as a way to preserve Khmer culture and dance for the future, and those that want to learn, or envision the aesthetics of Khmer culture in their projects.
Collaborators
Photography - Sokai Yoon
Tools - Stable Diffusion, Automatic1111, & Dreambooth