Genshin_LoRA

LoRA model of the Genshin Characters

Data source

Manually captured 64 characters at different angles in the character directory of the Genshin game, 20 pictures for each character. You can get data from RobertLau/Genshin.

Train

Training code: lora-scripts.

Batch training code: batch_train_loras.

Usage

prompt:

1[Gender], solo, [EN_Name], [Arbitrary scene description]

For example:

"1girl, solo, Nahida, on the playground

Negative prompt (You can choose the following three ways):

1. \
2. (((NSFW))),poorly drawn, anatomical nonsense, bad anatomy, bad feet, bad hands, bad proportions, quality
3. (worst quality, low quality:1.4), (monochrome:1.1), turn pale, text, watermark

from diffusers import StableDiffusionPipeline
import torch

model_id = "ckpt/anything-v3.0"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)

lora_file = "Nahida.safetensors"
pipe.load_lora_weights(lora_file)
pipe = pipe.to("cuda")

prompt = "1girl, Nahida, solo, on the playground"
image = pipe(prompt).images[0]
image.save("./hatsune_miku.png")

Result

Gender: girl
EN_Name: Nahida, YaeMiko, Ganyu
Arbitrary scene description: on the playground, in the classroom