Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,39 @@
|
|
1 |
---
|
2 |
license: creativeml-openrail-m
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: creativeml-openrail-m
|
3 |
---
|
4 |
+
|
5 |
+
# Okingjo's Single-identifier LORAs
|
6 |
+
|
7 |
+
I will share most of my LORA model with single identifier here. by saying "single", only one charater with one costume are stored within the model.
|
8 |
+
Not only will the LORA model will be post here, training setups and tips will also be shared.
|
9 |
+
I`m still in the state of learning, so any comments/feedbacks are welcom!
|
10 |
+
|
11 |
+
|
12 |
+
## Characters from Genshin Impact
|
13 |
+
|
14 |
+
### sangonomiya-kokomi / 珊瑚宫心海
|
15 |
+
#### Brief intro
|
16 |
+
LORA of Sangonomiya Kokomi, with her default costume in game.
|
17 |
+
civitAI page [download](https://civitai.com/models/9186/sangonomiya-kokomi)
|
18 |
+
#### Training dataset
|
19 |
+
149 images of Kokomi:
|
20 |
+
* 4 nude illustrations, to ensure the AI knows that the costume is removable
|
21 |
+
* 85 normal illustrations of Kokomi, multiple angle, style and composition
|
22 |
+
* 30 nude 360 degree snapshot of Kokomi's 3D model
|
23 |
+
* 30 normal 360 degree snapshot of Kokomi's 3D model
|
24 |
+
Since only one costume is included, all 149 images are placed inside one folder.
|
25 |
+
#### Captioning
|
26 |
+
WD14 captioning instead of the danbooru caption was used, since the former one will not crop/resize the images.
|
27 |
+
Threshold are usually set to 0.75-0.8. since I don't like to have a very long and sometimes inaccurate caption for my training data.
|
28 |
+
After captionin is done, I added "sangonomiya kokomi" after "1girl" to every caption file generate as the triggering prompt. Some of the caption files were empty so I have to manually type the words.
|
29 |
+
#### Training setup
|
30 |
+
Trained with Kohya_SS stable diffusion trainer
|
31 |
+
Base model was [Anything V3.0 full](https://huggingface.co/Linaqruf/anything-v3.0/blob/main/anything-v3-fp32-pruned.safetensors)
|
32 |
+
Trainig process consist of two phases. The first one with default parameters of:
|
33 |
+
* learning_rate: 0.0001
|
34 |
+
* text_encoder_lr: 5e-5
|
35 |
+
* unet_lr: 0.0001
|
36 |
+
20 repeats, and 5 epoch
|
37 |
+
Then, for phase2, all three learning rate were decreased to 1/10, and trained with another 5 epochs.
|
38 |
+
|
39 |
+
#### results
|