The text was generated by GPT.

文案由 GPT 生成

This model strictly prohibits reproduction, including but not limited to liblib and seaart.

该模型严禁转载，包括但不限于liblib和seaart。

This model is intended solely for learning and communication purposes and is not to be used for any commercial purposes.

该模型仅用于学习和交流，禁止用于任何商业用途。

The training of this model utilized two A40-48g graphics cards and underwent 300 hours of training.

训练本模型使用了两张A40-48g显卡，并历经300小时的训练。

We express our gratitude to WILLY for providing technical support.

我们感谢WILLY提供的技术支持。

It is recommended that the resolution of generated images not be lower than 832x1280, and it is strongly advised against using 512x768. It is recommended to use 1024x1536 instead.

推荐生成图像的分辨率不低于832x1280，强烈建议不要使用512x768，推荐使用1024x1536。

The intensity of high-definition restoration and redraw is not recommended to exceed 0.4.

高清修复重绘强度不建议超过0.4。

Quality word labels: masterpiece, best quality, high quality, normal quality, low quality, worst quality

质量词标签：masterpiece, best quality, high quality, normal quality, low quality, worst quality

Recommended negative embeddings: unaestheticXLv13, negativeXL_D

推荐的负面嵌入：unaestheticXLv13unaestheticXL | Negative TI - v1.3 | Stable Diffusion Embedding | Civitai, negativeXL_DnegativeXL - D | Stable Diffusion Embedding | Civitai

pretrained_model训练底模【SDXL】/动漫/bulldozer_BETA - v2.0 |稳定扩散检查点 |奇维泰 (civitai.com)

我的常用负面提示词：unaestheticXLv13, negativeXL_D, bad anatomy, blurry, disembodied limb, Two navel eyes, worst quality, Hand with excessive fingers,Foot with excessive toes,Hand with surplus fingers,Hand with deficient fingers,Abnormal anatomy,Poorly formed hands,Deformed hands and fingers,Additional limbs,Excessive limbs,Abnormal finger interlocking duplicate, cropped, text, jpeg, artifacts, signature, watermark, username, blurry, artist name, trademark, title, muscular, sd character, multiple view, Reference sheet, long body, malformed limbs, multiple breasts, cloned face, malformed, mutated, bad anatomy, disfigured, bad proportions, duplicate, bad feet, artist name, extra limbs, ugly, fused anus, text font ui, missing, bad foot quality, twisted toes, asymmetric toes, abnormal toe shapes, misplaced toes, missing toes, extra toes, oversized/small toes, misaligned ankles, mismatched foot sizes, scars or blemishes, nipple-like toes, swollen or deformed toes, overlapping toes, excess toenail, incorrect toe count, uneven toe count, unnatural foot proportions.

The training parameters are as follows:

训练参数如下

resolution = "1024,1024"

enable_bucket = true

min_bucket_reso = 256

max_bucket_reso = 1_280

bucket_reso_steps = 32

output_name = "XLhans7"

output_dir = "/root/autodl-tmp"

save_model_as = "safetensors"

save_precision = "fp16"

save_every_n_epochs = 10

max_train_epochs = 200

train_batch_size = 20

gradient_checkpointing = true

gradient_accumulation_steps = 1

learning_rate = 0.00001

learning_rate_te = 5e-7.0

lr_scheduler = "cosine_with_restarts"

lr_warmup_steps = 0

lr_scheduler_num_cycles = 1

optimizer_type = "AdamW8bit"

log_with = "tensorboard"

logging_dir = "./logs"

caption_extension = ".txt"

shuffle_caption = true

weighted_captions = false

keep_tokens = 8

max_token_length = 255

multires_noise_iterations = 6

multires_noise_discount = 0.4

seed = 1_337

clip_skip = 2

no_token_padding = false

mixed_precision = "bf16"

full_fp16 = false

full_bf16 = true

xformers = true

lowram = false

cache_latents = true

cache_latents_to_disk = true

persistent_data_loader_workers = true

multi_gpu = false

gpu_ids = [ "GPU 0: NVIDIA A40 (47 GB)", "GPU 1: NVIDIA A40 (47 GB)" ]

This model strictly prohibits any commercial use.

该模型严禁一切商业使用。

The addition of the Artist prefix and Style suffix has minimal impact on the generated results and may be omitted.

添加Artist的前缀和 Style的后缀对生成结果影响很小，可以不添加。

个人原因部分艺术家名字和danbooru完全不同

（这里包含了艺术家的列表**艺术家（Artist）：
1. omone hokoma agm
2. sheya
3. yunkaiming
4. hennnachoco
5. chen bin
6. hishikihishiki (pixi14719710)
7. jszhdzcyz
8. silveroid
9. akmkmk3
10. 遙（yao）
11. hagi (ame hagi)
12. ogipote
13. wakaba
14. miwano rag
15. u_u_zan
16. hong bai
17. hokori sakuni
18. yonemuro
19. shiratama (shiratamaco)
20. touzai (poppin phl95)
21. mochizuki rik
22. miyase_mahiro
23. tianliang duohe fangdongye
24. catsmoon
25. kacyu
26. shengtian
27. kase daiki
28. HAPPOBIJIN
29. gyeou
30. ru zhai
31. ito lab
32. zhongwu chahui
33. rin yuu
34. milkshakework
35. myowa
36. fukahire_(ruinon)
37. funii
38. myush
39. iuui
40. timitarcat
41. sutorora
42. momoko (momopoco)
43. houkisei
44. himitsu_(hi_mi_tsu_2)
45. tadano_yukiko
46. tyakomes
47. shirabi
48. classic (zildjian33)
49. kimishima_ao
50. onineko
51. ameto_yuki
52. haru (hiyori-kohal)
53. peachyp
54. z282g
55. azuuru
56. wlop
57. Xnkzzmcz
58. Ynnnxs
59. weri
60. void0
61. vtihmdkdmfty
62. Unfairr
63. Vhmd
64. sousouman
65. torino_aqua
66. tinker
67. tokky
68. SHC
69. siki
70. sakura
71. say_hana
72. RailgunKy
73. roha
74. rokusai
75. Rta
76. quan (kurisu tina)
77. qiandaiyiyu
78. pottsness
79. oyuwari
80. na-ga
81. NNmy_Itp0
82. m-da s-tarou
83. momomoi momoi
84. msh
85. Miv4t
86. misyune
87. Machi馬吉
88. mafuyu (chibi21)
89. Ltt
90. lhlmcdm
91. Koahri
92. Leviathan
93. kisaragi yaya
94. kataokasan
95. ke-ta
96. iumu
97. IorI
98. hiten_(hitenkei)
99. Hitoimim
100. haneru
101. hanekoto
102. Egami
103. GiO
104. GongHa
105. dsmile
106. DIno
107. Cqrm
108. deyui
109. atdan
110. BPG
111. bzmfbgdm
112. Nachoneko
113. Anmi
114. Archive
115. alpha91
116. amane utata
117. jima

**风格 (Style):**
1. Blue Archive Style
2. best quality Style
3. Azur Lane Style
4. uzu-Soft Style
5. arknights Style
6. game cg Style
7. Touhou lostword Style
8. mmd Style
9. SD1.5 Style
10. 3D Style
11. 2023 Era Style
12. 1999 Style
13. nai3 Style
14. Thick Painting Style