Create release_notes/v1.1-hisakawa.md
Browse files
hisakawa/release_notes/v1.1-hisakawa.md
ADDED
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# v1.1-hisakawa
|
2 |
+
|
3 |
+
A LoRA (Hadamard Product) resumed from [deresute-v1.1](https://civitai.com/models/23476/the-idolmster-cinderella-girls-starlight-stage-style-90-characters) and tuned on 72 Starlight Stage cards and character sprites, improving consistency and likeness of the twins while retaining the ability to recognize other characters.
|
4 |
+
|
5 |
+
This is not a style-neutral model. Currently, generating both characters in the same image isn't supported. This serves as a preview model while these issues are being worked on.
|
6 |
+
|
7 |
+
This shows that official material is sufficient and selective finetuning is possible.
|
8 |
+
|
9 |
+
If there are other characters whose likeness you find that the [base model](https://civitai.com/models/23476/the-idolmster-cinderella-girls-starlight-stage-style-90-characters) doesn't capture well, please leave a comment below. The feedback would help improve the base model.
|
10 |
+
|
11 |
+
Usage
|
12 |
+
-----
|
13 |
+
|
14 |
+
This Hadamard Product LoRA requires installing [an extra extension](https://github.com/KohakuBlueleaf/a1111-sd-webui-locon) to be used in the Web UI.
|
15 |
+
|
16 |
+
The style itself doesn't have a trigger word. For the characters, below are the character tags and respective top 20 related tags.
|
17 |
+
|
18 |
+
```
|
19 |
+
Hisakawa Nagi,
|
20 |
+
braid, 1girl, grey hair, brown eyes, long hair, twintails, low twintails, bangs, solo, looking at viewer, shirt, braided bangs, white background, simple background, long sleeves, skirt, shorts, thighhighs, holding, hair ribbon
|
21 |
+
|
22 |
+
Hisakawa Hayate,
|
23 |
+
long hair, grey hair, bangs, braid, 1girl, jewelry, solo, earrings, blue eyes, looking at viewer, braided bangs, smile, blush, very long hair, shirt, white background, simple background, skirt, breasts, long sleeves
|
24 |
+
```
|
25 |
+
|
26 |
+
As this is a finetuned model, prompts that work on the [base model](https://civitai.com/models/23476/the-idolmster-cinderella-girls-starlight-stage-style-90-characters) should also work.
|
27 |
+
|
28 |
+
Training info
|
29 |
+
-------------
|
30 |
+
|
31 |
+
Finetuned (without optimizer state) from [deresute-v1.1](https://civitai.com/models/23476/the-idolmster-cinderella-girls-starlight-stage-style-90-characters) with resolution 768*768.
|
32 |
+
|
33 |
+
Training cost: ~1 T4-hour
|
34 |
+
|
35 |
+
Training config:
|
36 |
+
|
37 |
+
```
|
38 |
+
[model_arguments]
|
39 |
+
v2 = false
|
40 |
+
v_parameterization = false
|
41 |
+
pretrained_model_name_or_path = "Animefull-final-pruned.ckpt"
|
42 |
+
|
43 |
+
[additional_network_arguments]
|
44 |
+
no_metadata = false
|
45 |
+
network_module = "lycoris.kohya"
|
46 |
+
network_dim = 16
|
47 |
+
network_alpha = 1
|
48 |
+
network_args = [ "conv_dim=0", "conv_alpha=16", "algo=loha",]
|
49 |
+
network_train_unet_only = false
|
50 |
+
network_train_text_encoder_only = false
|
51 |
+
|
52 |
+
[optimizer_arguments]
|
53 |
+
optimizer_type = "AdamW8bit"
|
54 |
+
learning_rate = 1e-4
|
55 |
+
lr_scheduler = "cosine"
|
56 |
+
lr_warmup_steps = 80
|
57 |
+
|
58 |
+
[dataset_arguments]
|
59 |
+
debug_dataset = false
|
60 |
+
|
61 |
+
[training_arguments]
|
62 |
+
save_precision = "fp16"
|
63 |
+
save_every_n_epochs = 1
|
64 |
+
train_batch_size = 2
|
65 |
+
max_token_length = 225
|
66 |
+
mem_eff_attn = false
|
67 |
+
xformers = true
|
68 |
+
max_train_epochs = 40
|
69 |
+
max_data_loader_n_workers = 8
|
70 |
+
persistent_data_loader_workers = true
|
71 |
+
gradient_checkpointing = false
|
72 |
+
gradient_accumulation_steps = 1
|
73 |
+
mixed_precision = "fp16"
|
74 |
+
clip_skip = 2
|
75 |
+
lowram = true
|
76 |
+
|
77 |
+
[saving_arguments]
|
78 |
+
save_model_as = "safetensors"
|
79 |
+
```
|