This model is finetune from Katherine Crowson's fine-tuned 512x512 diffusion model (https://github.com/openai/guided-diffusion), using 1k+ nature image crawled from the InterNet.