Stable Diffusion Regularization Images
All classes in Stable Diffusion 1.5 and 2.1 checkpoint
Quantity of images: 5k per class
Images generated with following parameters:
- Classification CFG Scale: 7,5
- Steps: 50
- Scheduler: ddim
- Resolution: 768px (2.1) and 512px (1.5)
- Prompt: "photo of a {}"
Done:
- photo of a person (1.5);
- photo of a person (2.1);
- photo of a man (1.5);
- photo of a man (2.1);
- photo of a woman (1.5);
- photo of a woman (2.1).
Todo:
- maybe use negative prompt on images generated in SD 2.1 (checking if this will affect the final result).
suggestions? open a issue