Training Process

• Epochs: 15

• Repeats: 3

• Dataset Size: 136 images

Training Method

Due to the complete lack of available dataset images, I conducted training in two phases:

• First Phase: Primarily trained character appearance by capturing two types of character illustrations with different skins from the game (approximately 30 images)

• Second Phase: Used the model trained in the first phase, enlarged the continuous animations output from Spine, and then refined them

As a result, this Lora model can generally reproduce any motion that Spine can create.

Trigger Words

Skin 1: lancelot, lancelot_skin1, hair_ornament, mole_under_mouth, red_lipstick, pantyhose

Skin 2: lancelot, lancelot_skin2, hair_ornament, navel, swimsuit, flower, hair_flower, huge_breasts, mole_under_mouth, slingshot_swimsuit