Trained om mostly small objects and clothing so some things might need clever prompt engineering.
Description
Better captioning, less steps (v1.0 is overtrained) and a few more images.
Works better with most checkpoints now.
Thanks to @ZombieHead for making me change captioning method.