Fixed a couple things here in the training data, upped the dim to capture more reliable fangs, and used "teeth" as a base class, which I think makes for more "human" fangs and not fangs like animals, which is what the 1.5 model was trained on (and why I would end up with all teeth being fangs, or weirdly misplaced ones).
Shout-out to Rusher7 who I coincidentally found out was also training a fang Lora, and who I bounced some ideas off of this month.
Fixed a couple things here in the training data, upped the dim to capture more reliable fangs, and used "teeth" as a base class, which I think makes for more "human" fangs and not fangs like animals, which is what the 1.5 model was trained on (and why I would end up with all teeth being fangs, or weirdly misplaced ones).
Shout-out to Rusher7 who I coincidentally found out was also training a fang Lora, and who I bounced some ideas off of this month.