khanon commited on
Commit
cca4c1c
1 Parent(s): a0d2fa7

updates Seia LoRA to v1b

Browse files
Files changed (2) hide show
  1. .gitattributes +1 -0
  2. seia/README.md +12 -3
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zst filter=lfs diff=lfs merge=lfs -text
34
  *tfevents* filter=lfs diff=lfs merge=lfs -text
35
  *.png filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zst filter=lfs diff=lfs merge=lfs -text
34
  *tfevents* filter=lfs diff=lfs merge=lfs -text
35
  *.png filter=lfs diff=lfs merge=lfs -text
36
+ *.jpg filter=lfs diff=lfs merge=lfs -text
seia/README.md CHANGED
@@ -1,7 +1,7 @@
1
  # Yurizono Seia (Blue Archive)
2
  百合園セイア (ブルーアーカイブ) / 유리조노 세이아 (블루 아카이브) / 百合園圣娅 (蔚藍檔案)
3
 
4
- [**Download here.**](https://huggingface.co/khanon/lora-training/blob/main/seia/chara-seia-v1.safetensors)
5
 
6
  ## Table of Contents
7
  - [Preview](#preview)
@@ -11,8 +11,8 @@
11
 
12
  ## Preview
13
  ![Seia portrait](chara-seia-v1.png)
14
- ![Seia preview 1](example-001-DefmixRed-v1.png)
15
- ![Seia preview 2](example-003-DefmixRed-v1.png)
16
 
17
  ## Usage
18
  Use any or all of the following tags to summon Arona: `seia, 1girl, fox ears, parted bangs, halo, multicolored eyes, flower wreath, blonde hair`
@@ -37,10 +37,19 @@ For her expressions: `expressionless` / `light smile` / `parted lips`
37
  - 1.5e-4 unet LR
38
  - 1e-5 optimizer LR
39
  - Used network_dimension 128 (same as usual) / network alpha 128 (default)
 
40
  - Training resolution 832x832.
41
  - Trained without VAE.
42
  - [Dataset can be found on the mega.co.nz repository.](https://mega.nz/folder/SupXFCiC#C992OYnVa8svZt3PdUObww)
43
 
44
  ## Revisions
 
 
 
 
 
 
 
 
45
  - v1 (2023-02-12)
46
  - Initial release.
 
1
  # Yurizono Seia (Blue Archive)
2
  百合園セイア (ブルーアーカイブ) / 유리조노 세이아 (블루 아카이브) / 百合園圣娅 (蔚藍檔案)
3
 
4
+ [**Download here.**](https://huggingface.co/khanon/lora-training/blob/main/seia/chara-seia-v1b.safetensors)
5
 
6
  ## Table of Contents
7
  - [Preview](#preview)
 
11
 
12
  ## Preview
13
  ![Seia portrait](chara-seia-v1.png)
14
+ ![Seia preview 1](example-001-DefmixRed-v1b.png)
15
+ ![Seia preview 2](example-003-DefmixRed-v1b.png)
16
 
17
  ## Usage
18
  Use any or all of the following tags to summon Arona: `seia, 1girl, fox ears, parted bangs, halo, multicolored eyes, flower wreath, blonde hair`
 
37
  - 1.5e-4 unet LR
38
  - 1e-5 optimizer LR
39
  - Used network_dimension 128 (same as usual) / network alpha 128 (default)
40
+ - Resized to 32 after training
41
  - Training resolution 832x832.
42
  - Trained without VAE.
43
  - [Dataset can be found on the mega.co.nz repository.](https://mega.nz/folder/SupXFCiC#C992OYnVa8svZt3PdUObww)
44
 
45
  ## Revisions
46
+ - v1b (2023-02-12)
47
+ - Resized v1 from network rank 128 >>> 32. This has no negative impact on image quality/coherency, and provides these advantages:
48
+ - Reduced LoRA file size from 144mb to 36mb
49
+ - Slightly reduced overfitting on training data
50
+ - Less darkening/vignetting of backgrounds
51
+ - Improved color saturation
52
+ - Halo is a bit more consistent
53
+ - Seia's hair gets confused with details on her clothing less frequently
54
  - v1 (2023-02-12)
55
  - Initial release.