In order to create the images below, I went and visited some old Xenodream creations and prepared them as input images for another Dreambooth training.
The objects are made using spherical displacement driven by various mathematical equations. In addition, complexity and size can be altered.
In total, I gathered 18 images and cropped them to 512 by 512 pixels. After an hour of crunching, the custom Stable Diffusion model was ready.
I created a list of prompts and had Stable Diffusion GUI iterate through them overnight. I woke up with 1200 fresh images, and selected 40 top results. Enjoy!