LoRA for Karis from Golden Sun: Dark Dawn. She's cute, so I thought she needed a LoRA so people could have her do cute things. Please upload your pictures you make with this, even if they're failures.
V2 Prompts
Trigger: Karis_GS
Prompt: green hair, ponytail, bangs, asymmetrical hair, sidelocks, purple eyes, triangle earrings, capelet, blue capelet, small breasts, mandarin collar, coat, white coat, ribbon trim, gold trim, undershirt, pink shirt, pink sleeves, crossed belts, miniskirt, pink skirt, white gloves, elbow gloves, bracelet, black pantyhose, white boots, knee boots
Other tags in data you might be interested in: fighting stance, magic, holding, holding weapon, weapon, staff, sword
Negatives: monochrome, greyscale, manga illustration, pixel art, speech bubble, blank speech bubble, 3d, low poly (all tags used in training data that are undesirable for most generations)
(for V1’s prompts, see the version info)
V2 Adds 11 new fanart pieces to bring total images to 32 (on top of the original’s 9 pictures from the promo manga, 3 pieces of official art, 4 screenshots of her in game model, a screenshot of her chibi model, her in-game portrait, and 3 pieces of fan art). I made the original when I was still figuring out how many pics were optimal, didn’t realize there were a bunch of images of her that were just tagged with the series and not her name on Pixiv, wasn’t as confident in my cleaning ability, and didn't have as good a handle on how to tag stuff. The improvements on V2 made it train very quickly and be accurate by epoch 2.
Description
LoRA for Karis from Golden Sun: Dark Dawn trained on 21 images (consisting of 9 pictures from the promo manga, 3 pieces of official art, 4 screenshots of her in game model, a screenshot of her chibi model, her in-game portrait, and 3 pieces of fan art). She's cute, so I thought she needed a LoRA so people could have her do cute things. I wonder how this turns out given the sub-optimal training data: Please upload your pictures you make with this, even if they're failures. Edit: This came out pretty well, just a little stiff if not paired with other resources (I still want to see other people's pics).
Trigger: Karis_GS
Prompt: 1girl, green hair, ponytail, purple eyes, earrings, blue capelet, small breasts, mandarin collar, white coat, ribbon trim, gold trim, crossed belts, pink miniskirt, pink undershirt, pink sleeves, white gloves, elbow gloves, bracelet, black pantyhose, white boots
Other tags in data you might be interested in: fighting stance, magic, holding, holding weapon, weapon, staff, sword (her in-game model, holding a short sword, is almost 1/5th of the training data, while almost half the data has her wielding a staff.)
Negatives: monochrome, greyscale, manga illustration, pixel art, speech bubble, blank speech bubble, 3d, game screenshot, low poly (all tags used in training data that are undesirable for most generations)