>>37317837the base stable diffusion model was trained on millions to billions of images crawled from the internet, and the big models in use just have additional anime training on top of that, so the model has at least a slight understanding of a surprising variety of concepts
