gustproof commited on
Commit
a0be49c
1 Parent(s): aa6a183

Create v1.0.md

Browse files
Files changed (1) hide show
  1. deresute-u149/release_notes/v1.0.md +170 -0
deresute-u149/release_notes/v1.0.md ADDED
@@ -0,0 +1,170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ A LoRA (Hadamard Product) based on Cinderella Girls Cards and Starlight Stage Cards of the [9 characters in the upcoming U149 anime](https://cinderella-u149-anime.idolmaster-official.jp/character). A small test before training for many more characters in Starlight Stage.
2
+
3
+ **Note**: the site ~~does not display~~ discards txt2img parameters with non-ASCII characters. Below are the generation parameters of the preview images:
4
+
5
+ ```
6
+ 橘ありす,
7
+ 1girl, :o, apron, blue dress, blush, brown hair, card, dress, fiatto029, flower, half updo, jewelry, key necklace, long hair, looking at viewer, necklace, pinafore dress, playing card, purple flower, purple rose, rose, round teeth, solo, teeth, very long hair, white apron, wrist cuffs
8
+ <lora:deremas-lite-v1-linear:1>
9
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 1406663243, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
10
+
11
+ 櫻井桃華,
12
+ 1girl, :d, arm up, bangs, blonde hair, blurry, blurry background, blush, cowboy shot, flower, green eyes, hair flower, hair ornament, long sleeves, looking at viewer, open mouth, short hair, smile, solo, white flower, white dress
13
+ <lora:deremas-lite-v1-linear:1>
14
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 3140247075, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
15
+
16
+ 赤城みりあ,
17
+ 1girl, black hair, blush, floral print, flower, hair flower, hair ornament, japanese clothes, kimono, looking at viewer, own hands together, portrait, simple background, smile, solo, two side up, white background, white kimono
18
+ <lora:deremas-lite-v1-linear:1>
19
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 2937053596, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
20
+
21
+ 的場梨沙,
22
+ 1girl, animal print, belt, bow, brown hair, crop top, cropped jacket, hair between eyes, hair bow, idol, jacket, jewelry, leopard print, midriff, miniskirt, nail polish, navel, necklace, one eye closed, open mouth, pink nails, pink skirt, pleated skirt, sidelocks, skirt, smile, solo, twintails, white jacket, yellow eyes
23
+ <lora:deremas-lite-v1-linear:1>
24
+ Negative prompt: worst quality
25
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 186615327, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
26
+
27
+ 結城晴,
28
+ 1girl, bag, bangs, bike shorts, black jacket, black shorts, blazer, brown eyes, collared shirt, day, dress shirt, grey background, grey skirt, grin, hair between eyes, head tilt, jacket, long hair, long sleeves, looking at viewer, necktie, open clothes, open jacket, orange hair, plaid, plaid skirt, pleated skirt, red necktie, school bag, school uniform, shirt, short shorts, shorts, shorts under skirt, skirt, smile, solo, white shirt
29
+ <lora:deremas-lite-v1-linear:1>
30
+ Negative prompt: worst quality
31
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 2606433895, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
32
+
33
+ 佐々木千枝,
34
+ 1girl, black eyes, black hair, blush, braid, bridal veil, dress, flower basket, flower ornament, glint, gloves, jewelry,, lace-trimmed veil, lace trim, looking at viewer, ornate ring, ring, sleeveless, sleeveless dress, smile, solo, veil, wedding dress, wedding ring, white dress, white gloves
35
+ <lora:deremas-lite-v1-linear:1>
36
+ Negative prompt: worst quality
37
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 1209301241, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
38
+
39
+ 龍崎薫,
40
+ 1girl, :o, blue sky, blush, brown eyes, brown hair, cardigan, cloud, collarbone, from below, green cardigan, hair ornament, hairclip, hood, hood down, hooded cardigan, looking at viewer, open cardigan, open clothes, open mouth, red shirt, shirt, short hair, sky, solo
41
+ <lora:deremas-lite-v1-linear:1>
42
+ Negative prompt: worst quality
43
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 1738586102, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
44
+
45
+ 市原仁奈,
46
+ 1girl, brown eyes, brown hair, dancing, hat, long hair, open mouth, outstretched arms, outstretched hand, pink headwear, pink shirt, pink skirt, plaid headwear, sailor collar, shirt, skirt, smile, solo, suspender skirt, suspenders, television
47
+ <lora:deremas-lite-v1-linear:1>
48
+ Negative prompt: worst quality
49
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 3419865117, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
50
+
51
+ 古賀小春,
52
+ 1girl, animal, arm up, bag, bangs, bare shoulders, bench, bird, blush, brown eyes, brown hair, brown sleeves, bush, closed mouth, collarbone, day, detached sleeves, dress, falling leaves, frilled dress, frills, hair between eyes, hand on headwear, hat, leaf, long sleeves, outdoors, sleeveless, sleeveless dress, sleeves past wrists, smile, solo, sun hat, sundress, tree, white headwear, yellow dress
53
+ <lora:deremas-lite-v1-linear:1>
54
+ Negative prompt: worst quality
55
+ Steps: 20, Sampler: DPM++ 2S a Karras, CFG scale: 5, Seed: 4091483865, Size: 512x768, Model: AOM3A1_orangemixs, Clip skip: 2
56
+ ```
57
+
58
+ Usage
59
+ -----
60
+
61
+ This Hadamard Product LoRA requires installing [an extra extension](https://github.com/KohakuBlueleaf/a1111-sd-webui-locon) to be used in the Web UI.
62
+
63
+ The style itself doesn't have a trigger word. For the characters, below are the character tags and respective top 20 related tags.
64
+
65
+ ```
66
+ 古賀小春,
67
+ short hair, brown hair, 1girl, smile, brown eyes, solo, open mouth, bow, dress, looking at viewer, hair bow, hairband, blush, sitting, frills, light brown hair, sky, :d, gloves, blonde hair
68
+
69
+ 赤城みりあ,
70
+ short hair, black hair, smile, brown eyes, two side up, open mouth, looking at viewer, 1girl, solo, blush, hair ornament, :d, brown hair, multiple girls, skirt, jewelry, one eye closed, closed eyes, gloves, shorts
71
+
72
+ 結城晴,
73
+ orange hair, smile, 1girl, solo, long hair, open mouth, shorts, looking at viewer, blush, sky, day, brown eyes, gloves, outdoors, multiple girls, brown hair, midriff, navel, hat, shirt
74
+
75
+ 龍崎薫,
76
+ short hair, hair ornament, hairclip, smile, brown hair, open mouth, 1girl, yellow eyes, solo, orange hair, shorts, looking at viewer, closed eyes, hat, brown eyes, blush, food, multiple girls, flower, skirt
77
+
78
+ 的場梨沙,
79
+ black hair, long hair, yellow eyes, twintails, smile, jewelry, 1girl, ribbon, solo, hair ribbon, open mouth, necklace, looking at viewer, blush, skirt, animal print, bangs, midriff, one eye closed, sitting
80
+
81
+ 橘ありす,
82
+ brown eyes, brown hair, long hair, 1girl, solo, bow, looking at viewer, smile, hair bow, dress, open mouth, blush, ribbon, food, flower, sitting, fruit, jewelry, hair ornament, strawberry
83
+
84
+ 佐々木千枝,
85
+ short hair, black hair, smile, black eyes, hair ornament, 1girl, rabbit hair ornament, solo, open mouth, hairclip, looking at viewer, blush, skirt, dress, brown hair, jewelry, holding, multiple girls, sitting, indoors
86
+
87
+ 櫻井桃華,
88
+ blonde hair, green eyes, short hair, smile, 1girl, looking at viewer, solo, blush, wavy hair, dress, open mouth, frills, hairband, flower, hair ornament, bow, gloves, jewelry, rose, sitting
89
+
90
+ 市原仁奈,
91
+ long hair, brown hair, smile, open mouth, blush, 1girl, brown eyes, solo, blunt bangs, cosplay, animal costume, kigurumi, :d, bow, bangs, looking at viewer, closed eyes, gloves, outdoors, hat
92
+ ```
93
+
94
+ Training info
95
+ -------------
96
+
97
+ Trained solely on 164 640x800 frameless Cinderella Girls Cards and 273 1280x824 Starlight Stage Cards collected from [アイマス シンデレラガールズ カードギャラリー](https://imas.gamedbs.jp/cg/) without balancing and tag cleaning. Near-duplicates were found (minor differences like closed/open eyes), but they were not cleaned. As the jpgs are highly compressed, they are passed through [RealESRGAN_x4plus_anime_6B](https://github.com/xinntao/Real-ESRGAN/blob/master/docs/anime_model.md) with --outscale=1 in order to remove the artifacts. Tags generated with [SmilingWolf/wd-v1-4-swinv2-tagger-v2](https://huggingface.co/SmilingWolf/wd-v1-4-swinv2-tagger-v2) with a threshold of 0.35. Caption template was "[char_name], [filewords]".
98
+
99
+ Originally trained with LoRA on convolution blocks ("LoCon"), but it was found that removing the convolution weights from the output doesn't alter the results significantly. The original is available as `v1-full`
100
+
101
+ and the conv-pruned version is `v1-linear`
102
+
103
+ Training cost: ~4.5 T4-hours
104
+
105
+ Data distribution:
106
+
107
+ ```
108
+ Character Total 640 1280
109
+ ---------------------------------
110
+ 櫻井桃華 60 36 24
111
+ 市原仁奈 56 38 18
112
+ 橘ありす 54 34 20
113
+ 赤城みりあ 53 37 16
114
+ 佐々木千枝 52 30 22
115
+ 龍崎薫 50 34 16
116
+ 結城晴 44 26 18
117
+ 的場梨沙 40 22 18
118
+ 古賀小春 28 16 12
119
+ ```
120
+
121
+ LoRA meta dump:
122
+
123
+ ```
124
+ {'ss_cache_latents': 'False',
125
+ 'ss_caption_dropout_every_n_epochs': '0',
126
+ 'ss_caption_dropout_rate': '0.0',
127
+ 'ss_caption_tag_dropout_rate': '0.0',
128
+ 'ss_clip_skip': '2',
129
+ 'ss_dataset_dirs': '{"meta_lat.json": {"n_repeats": 1, "img_count": 437}}',
130
+ 'ss_datasets': '[{"is_dreambooth": false, "batch_size_per_device": 3, "num_train_images": 437, "num_reg_images": 0, "resolution": [768, 768], "enable_bucket": true, "min_bucket_reso": null, "max_bucket_reso": null, "tag_frequency": {"meta_lat.json": {}}, "bucket_info": {"buckets": {"0": {"resolution": [640, 768], "count": 273}, "1": {"resolution": [896, 576], "count": 164}}, "mean_img_ar_error": 0.0}, "subsets": [{"img_count": 437, "num_repeats": 1, "color_aug": false, "flip_aug": false, "random_crop": false, "shuffle_caption": true, "keep_tokens": 0, "image_dir": "train_data", "metadata_file": "meta_lat.json"}]}]',
131
+ 'ss_epoch': '45',
132
+ 'ss_face_crop_aug_range': 'None',
133
+ 'ss_full_fp16': 'False',
134
+ 'ss_gradient_accumulation_steps': '1',
135
+ 'ss_gradient_checkpointing': 'False',
136
+ 'ss_learning_rate': '0.0001',
137
+ 'ss_lowram': 'True',
138
+ 'ss_lr_scheduler': 'polynomial',
139
+ 'ss_lr_warmup_steps': '100',
140
+ 'ss_max_grad_norm': '1.0',
141
+ 'ss_max_token_length': '225',
142
+ 'ss_max_train_steps': '7300',
143
+ 'ss_mixed_precision': 'fp16',
144
+ 'ss_network_alpha': '16',
145
+ 'ss_network_dim': '16',
146
+ 'ss_network_module': 'lycoris.kohya',
147
+ 'ss_new_sd_model_hash': '89d59c3dde4c56c6d5c41da34cc55ce479d93b4007046980934b14db71bdb2a8',
148
+ 'ss_noise_offset': 'None',
149
+ 'ss_num_batches_per_epoch': '146',
150
+ 'ss_num_epochs': '50',
151
+ 'ss_num_reg_images': '0',
152
+ 'ss_num_train_images': '437',
153
+ 'ss_optimizer': 'bitsandbytes.optim.adamw.AdamW8bit',
154
+ 'ss_output_name': 'u149',
155
+ 'ss_prior_loss_weight': '1.0',
156
+ 'ss_sd_model_hash': '925997e9',
157
+ 'ss_sd_model_name': 'Animefull-final-pruned.ckpt',
158
+ 'ss_sd_scripts_commit_hash': '5b506b14a3f0cda56c0012bb502590e785f5ff21',
159
+ 'ss_seed': 'None',
160
+ 'ss_session_id': '505064346',
161
+ 'ss_tag_frequency': '{"meta_lat.json": {}}',
162
+ 'ss_text_encoder_lr': '5e-05',
163
+ 'ss_training_comment': 'None',
164
+ 'ss_training_finished_at': '1678689998.9046714',
165
+ 'ss_training_started_at': '1678673431.1014886',
166
+ 'ss_unet_lr': '0.0001',
167
+ 'ss_v2': 'False',
168
+ 'sshs_legacy_hash': '65cdfd7c',
169
+ 'sshs_model_hash': '6a126d6984ae0903678f7a149e18a0b2ce34bf6e9ee4bb10311c3f1b82ea2901'}
170
+ ```