UrangDiffusion v2.5版本v2.0 (ID: 1002680)

UrangDiffusion v2.5 (oo-raw-ng Diffusion) brings a whole-new training method compared to the v2.0. The model provide more flexibility and brings some updated dataset.

The name “Urang” comes from Sundanese, meaning “We/Our/I.” The history behind the name is to make the model not only suitable for me but also for many people. Another reason is that I use many resources (training scripts, dataset collecting scripts, etc.) from other people. It’s unfair to claim this model as “my sole work.”

Standard Prompting Guidelines

The model is finetuned from Animagine XL 3.1, which is trained with danbooru tags. However, there is a little bit changes on dataset captioning, therefore there is some different default prompt used:

• Default prompt: 1girl/1boy, character name, from what series, everything else in any order, masterpiece, best quality, high score, great score, absurdres.

• Default negative prompt: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, low score, bad score, average score, signature, watermark, username, blurry,

• Default configuration: Euler a with around 25-30 steps, CFG 5-7, and ENSD set to 31337. Sweet spot is around 25 steps and CFG 7.

Training Configurations

Finetuned from: UrangDiffusion v2.0

Pretraining I (Crashed mid training):

• Dataset size: ~50,300 images

• GPU: 1xH100 80GB

• Optimizer: AdaFactor

• Unet Learning Rate: 3.75e-6

• Text Encoder Learning Rate: 1.875e-6

• Batch Size: 16

• Gradient Accumulation: 3

• Warmup steps: 100 steps

• Min SNR: 5

• Epoch: 4

• Random Cropping: True

• Loss: Huber

• Huber Schedule: SNR

• Huber C: 0.1

Pretraining II (Continue from epoch 4):

• Dataset size: ~50,300 images

• GPU: 1xA100 80GB

• Optimizer: AdaFactor

• Unet Learning Rate: 3.75e-6

• Text Encoder Learning Rate: 1.875e-6

• Batch Size: 16

• Gradient Accumulation: 3

• Warmup steps: 100 steps

• Min SNR: 5

• Epoch: 6

• Random Cropping: True

• Loss: Huber

• Huber Schedule: SNR

• Huber C: 0.1

CyberFix:

Epoch 10 + (Cyberrealistic XL v3.1 - SDXL 1.0 Base) = Temp

Temp model's text encoder changed with Epoch 10's text encoder.

Added/Updated Series, Characters, and Styles

v2.5

Artists:

1. ningen mame

2. ciloranko

3. rhasta

4. atdan

5. sho (sho lwlw)

6. tianliang

7. duohe

8. fangdongye

9. chen bin

10. ishikei

11. ask (askzy)

12. wlop

13. tsurusaki takahiro

14. kokosando

15. wagashi (dagashiya)

16. nawakena

17. kedama milk

18. hiten (hitenkei)

19. matanonki

20. sy4

21. houraku

22. fuzichoco

23. sencha (senchat)

24. rei (sanbonzakura)

25. houkisei

26. alp

27. dino (dinoartforame)

28. kuroduki (pieat)

29. maeda hiroyuki

30. tabi (tabisumika)

31. yuumei

32. rella

33. konya

34. karasue

35. hoshi (snacherubi)

36. modare

37. creayus

38. reoen

39. kawacy

40. wanke

41. kousaki rui

42. chyoel

43. lpip

44. kaninn

45. azuuru

46. mignon

47. amazuyu

48. tatsuki

49. shiro9jira

50. novelance

51. lack

52. airseal

53. huanxiang

54. heitu

55. rsef

56. machi (machi0910)

57. meion

58. z3zz4

59. ame (uten cancel)

60. healthyman

61. wagashi (dagashiya)

62. yamamoto souichirou

63. freng

64. kaede (sayappa)

65. masaki (ekakiningen)

66. asakuraf

67. misaka 12003-gou

68. hood (james x)

69. as109

70. yd (orange maru)

71. void 0

72. fajyobore

73. alphonse (white datura)

74. akita hika

75. nanaken

76. nana

77. muchi

78. maro

79. shisoneri

80. tottotonero

81. mochirong

82. nixeu

83. fujiyama

84. qizhu

85. kase daiki

86. ke-ta

87. tidsean

88. aki99

89. hitsukuya

90. shimmer

91. morikura en

92. ringeko-chan

93. pottsness

94. torino aqua

95. zelitto

96. personal ami

97. lm7 (op-center)

98. quan (kurisu tina)

99. migolu

100. shiki (psychedelic g2)

101. mizumizuni

102. kita (kitairoha)

103. kousaki rui

104. mofu

105. namako

106. omone

107. hokoma

108. agm

109. tab head

110. neoartcore

111. sciamano240

112. kuroboshi kouhaku

113. huke

114. lam (ramdayo)

115. nyori

116. yano mitsuki (nanairo)

117. yatomi

118. amashiro natsuki

119. yukisame

120. mishima kurone

121. teshima nari

122. shigure ui

123. orobou

v2.0

Series:

1. zenless zone zero

2. wuthering waves

3. sewayaki kitsune no senko-san

Honkai: Star Rail:

1. firefly

2. acheron

3. sparkle

4. robin

5. aventurine

6. black swan

7. feixiao

8. yunli

9. lingsha

10. march 7th (hunt)

11. jade

12. jiaoqiu

13. gallagher

14. rappa

15. misha

Hololive Talents:

1. hololive indonesia

2. raora panthera

3. elizabeth rose bloodflame

4. gigi murin

5. cecilia immergreen

Genshin Impact:

1. arlecchino

2. clorinde

3. chiori

4. mualani

5. xianyun

6. sigewinne

7. kinich

8. xilonen

9. emilie

10. gaming

11. kachina

12. sethos

Others:

1. landscape

2. several concepts to fix anatomy issue

Special Thanks

• My co-workers(?) at CagliostroLab for the insights and feedback.

• Nur Hikari and Vanilla Latte for quality control.

• Linaqruf, my tutor and role model in AI-generated images, and also the person behind tag ordering.

License

UrangDiffusion falls under the Fair AI Public License 1.0-SD license.

描述:

Using huber loss for better training.

Using random cropping for better image results.

Added new characters from new series, updated undertrained characters, and add new characters from existing series.

训练词语: