This folder contains models trained for the character grass wonder of umamusume.
The main goal is to test the difference between native training and lora.

You can prompt with 
- GrassWonder
- GrassWonder Toresen Uniform
- GrassWonder Racing Uniform

### Dataset

Totat size 376 (no regularization is used)
- Face crop: 149
- Racing uniform: 94
- Toresen uniform: 39
- Other grasswonder images: 94


### Base model

[NMFSAN](https://huggingface.co/Crosstyan/BPModel/blob/main/NMFSAN/README.md) so you can have different styles

### Native training

Trained with [Kohya trainer](https://github.com/Linaqruf/kohya-trainer)

- training of text encoder turned on
- learning rate 1e-6
- batch size 1
- clip skip 2
- number of training steps 7520 (20 epochs)

*Examples*
![native-00017-3566847929](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/native-00017-3566847929.png)
![native-00020-1546518749](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/native-00020-1546518749.png)
![native-00025-570458801](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/native-00025-570458801.png)


### LoRA

Please refer to [LoRA Training Guide](https://rentry.org/lora_train)

- training of text encoder turned on
- network dimension 128
- learning rate 1e-4
- batch size 6
- clip skip 2
- number of training steps 7520/6 (20 epochs)

*Examples*
![lora-00014-2366006784](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/lora-00014-2366006784.png)
![lora-00029-4052818765](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/lora-00029-4052818765.png)
![lora-00030-4052818765](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/lora-00030-4052818765.png)