gustproof commited on
Commit
7133b71
1 Parent(s): ae188c5

Create v2.0.md

Browse files
Files changed (1) hide show
  1. hisakawa/release_notes/v2.0.md +100 -0
hisakawa/release_notes/v2.0.md ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # v2.0
2
+
3
+ A LoRA of the Hisakawa twins from THE iDOLM@STER Cinderella Girls based on 967 fanarts. More style neutral than v1.1.
4
+
5
+ Two versions are available: rank 1 and rank 4.
6
+
7
+ This shows that, perhaps unsurprisingly, 1 rank can carry info of more than 1 character.
8
+
9
+ Despite the abundance of two-subject data in the training set, the model still struggles to seperate the attributes of the twins in the image. Increasing the model rank may or may not help. This will be investigated some time in the future.
10
+
11
+ ## Usage
12
+
13
+ The character tags are `Hisakawa Nagi` and `Hisakawa Hayate`.
14
+
15
+ To generate both characters in the same image, use either `Hisakawa Nagi / Hisakawa Hayate` or `Hisakawa Hayate / Hisakawa Nagi`
16
+
17
+
18
+ Top related tags:
19
+ ```
20
+ Hiwakawa Nagi,
21
+ 1girl, solo, twintails, long hair, grey hair, braid, low twintails, brown eyes, looking at viewer, ribbon, hair ribbon, braided bangs, blush, shirt, simple background, long sleeves, white background, black ribbon, very long hair, bow, white shirt, jacket, parted lips, skirt, collared shirt, closed mouth, thighhighs, hair over shoulder, :o, upper body, puffy sleeves, shorts, open mouth, breasts, red bow, black jacket, collarbone, short sleeves, holding, small breasts
22
+
23
+ Hisakawa Hayate,
24
+ 1girl, solo, long hair, grey hair, braid, looking at viewer, blush, smile, jewelry, breasts, braided bangs, earrings, blue eyes, very long hair, white background, shirt, open mouth, simple background, skirt, collarbone, medium breasts, white shirt, long sleeves, cleavage, pleated skirt, jacket, collared shirt, :d, bow, bare shoulders, navel, swimsuit, closed mouth, hair between eyes, holding, upper body, puffy sleeves, dress, bikini, short sleeves
25
+
26
+ Hisakawa Nagi / Hisakawa Hayate,
27
+ 2girls, multiple girls, siblings, twins, braid, long hair, sisters, grey hair, twintails, brown eyes, smile, braided bangs, open mouth, blue eyes, low twintails, ribbon, blush, looking at viewer, hair ribbon, jewelry, earrings, shirt, dress, skirt, long sleeves, black ribbon, very long hair, white shirt, bow, parted lips, breasts, :d, simple background, one eye closed, v, collarbone, white background, :o, puffy sleeves, collared shirt
28
+ ```
29
+
30
+ For specific outfits, refer to the preview images or the dataset.
31
+
32
+ ## Training info
33
+
34
+ Dataset: [967 fanarts from Danbooru](https://huggingface.co/datasets/gustproof/sd-data/blob/main/hisakawa.zip)
35
+ * 338 Hisakawa Nagi solo
36
+ * 341 Hisakawa Hayate solo
37
+ * 238 both twins
38
+
39
+ No balancing and tag cleaning.
40
+
41
+ For images with both characters, the order of the character tags were shuffled dynamically.
42
+
43
+
44
+ Training cost: ~2 T4-hour each for r1 and r4
45
+
46
+
47
+ Training config:
48
+ ```
49
+ [model_arguments]
50
+ v2 = false
51
+ v_parameterization = false
52
+ pretrained_model_name_or_path = "animefull-final-pruned.ckpt"
53
+
54
+ [additional_network_arguments]
55
+ no_metadata = false
56
+ unet_lr = 0.0001
57
+ text_encoder_lr = 0.0001
58
+ network_module = "networks.lora"
59
+ network_alpha = 1
60
+ network_train_unet_only = false
61
+ network_train_text_encoder_only = false
62
+
63
+ [optimizer_arguments]
64
+ min_snr_gamma = 5
65
+ optimizer_type = "AdamW8bit"
66
+ learning_rate = 0.0001
67
+ max_grad_norm = 1.0
68
+ lr_scheduler = "constant"
69
+ lr_warmup_steps = 0
70
+
71
+ [dataset_arguments]
72
+ debug_dataset = false
73
+ dataset_repeats = 1
74
+ shuffle_caption = true
75
+ keep_tokens = 1
76
+ resolution = "512,512"
77
+ caption_dropout_rate = 0
78
+ caption_tag_dropout_rate = 0
79
+ caption_dropout_every_n_epochs = 0
80
+ color_aug = false
81
+ token_warmup_min = 1
82
+ token_warmup_step = 0
83
+
84
+ [training_arguments]
85
+ save_precision = "fp16"
86
+ save_every_n_epochs = 1
87
+ train_batch_size = 4
88
+ max_token_length = 225
89
+ mem_eff_attn = false
90
+ xformers = true
91
+ max_train_epochs = 9999
92
+ max_data_loader_n_workers = 8
93
+ persistent_data_loader_workers = true
94
+ gradient_checkpointing = false
95
+ gradient_accumulation_steps = 1
96
+ mixed_precision = "fp16"
97
+ clip_skip = 2
98
+ lowram = true
99
+
100
+ ```