This was my first attempt at creating a separate Lycoris(LoHa) using the dataset from my CharHelper model + an extra hundred images. The biggest challenge was not having the mouth and teeth look like a janky mess. This model was the setup that gave me the best results. It is a LoHa trained with a 128 network rank 5 steps each for 22 epochs WITH 183 regularization images. The regularization images were important for achieving the closest likenesses.
Note: the model was not trained with any nudity, however, due to most orcs in the dataset being topless, a female Orc may show some nudity. To bypass this you can put "shirtless" or "topless" in the negative prompt.
Some useful keywords to activate the model are "CHV3COrc, Orc, Orcish, and Orc theme"
This was my first attempt at creating a separate Lycoris(LoHa) using the dataset from my CharHelper model + an extra hundred images. The biggest challenge was not having the mouth and teeth look like a janky mess. This model was the setup that gave me the best results. It is a LoHa trained with a 128 network rank 5 steps each for 22 epochs WITH 183 regularization images. The regularization images were important for achieving the closest likenesses.
Note: the model was not trained with any nudity, however, due to most orcs in the dataset being topless, a female Orc may show some nudity. To bypass this you can put "shirtless" or "topless" in the negative prompt.
Some useful keywords to activate the model are "CHV3COrc, Orc, Orcish, and Orc theme"