This Lora is not very precise, not sure if because of bad prompting or i just removed too many tags on training.
Also, it seems to prefer close-up shots, use upper body or cowboy shot tags.
Description
This time, it was trained with 1 repeat, 170 epochs, and without using the weight decay argument.
The quality seems to be a bit better, but it's still not very precise, probably it needs more images.