Sentinel7 commited on
Commit
9bfc24e
·
verified ·
1 Parent(s): 84b8963

Upload 2974434/3397856/README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. 2974434/3397856/README.md +142 -0
2974434/3397856/README.md ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ tags:
4
+ - anal
5
+ - anal insertion
6
+ - concept
7
+ - fisting
8
+ - gaping anus
9
+ - vagina
10
+ ---
11
+
12
+ # AnusPussy - AnusPussy Eros Ver v1.0 - 3397856
13
+
14
+ **Model Type**: LORA
15
+
16
+ **Base Model**: MiniMax H3
17
+
18
+ **Trigger Words**: zxqmei, zxqanus, zxqvagina
19
+
20
+ **Tags**: anal, anal insertion, concept, fisting, gaping anus, vagina
21
+
22
+ ## Gallery
23
+
24
+ <table>
25
+ <tr>
26
+ <td><video src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/dae5b948-b6cb-43c4-85a7-db46ff1c9e52/original=true/145305827.mp4" width="200" controls muted autoplay loop></video></td>
27
+ <td><video src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/175a8153-4f12-45d4-b127-f22960833f4e/original=true/145306136.mp4" width="200" controls muted autoplay loop></video></td>
28
+ <td><video src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8f6a9cbb-87eb-4181-92ff-2f076a73191a/original=true/145312730.mp4" width="200" controls muted autoplay loop></video></td>
29
+ </tr>
30
+ </table>
31
+
32
+ ## Description
33
+
34
+ [2026/10/11]
35
+
36
+ Upddate for MiniMax H3 EROS v5 CheckPoint int8 with base model to train lora ver.
37
+
38
+ I don't know why, but the Lora I trained with the offical checkpoint has very subtle results on EROS.
39
+
40
+ 更新基於MiniMax H3 EROS v5 CheckPoint int8 為基礎模型訓練的Lora版本
41
+
42
+ 不知道為什麼,我用官方checkpoint訓練的Lora在EROS 效果很微妙。
43
+
44
+ The data on the real people is still being collected slowly....
45
+
46
+ 真人的資料集還在緩慢收集中....
47
+
48
+ ===========================================================
49
+
50
+ Because testing MiniMax H3 shows that it often cannot clearly distinguish the positions of the anus and vagina, and does not understand anus-related NSFW actions.
51
+ (PS: It frequently confuses the anus and vagina, but in most cases it mistakes the anus for the vagina.)
52
+
53
+ Therefore I decided to train this LoRA myself. This is my first experimental LoRA.
54
+
55
+ All image materials are AI-generated. A total of 266 static images were used.
56
+
57
+ Generated by Stable Diffusion:
58
+
59
+ * Illustrious / Anima (about 90%)
60
+ * Partial 3D images (about 5%)
61
+ * Real photos (about 5%)
62
+
63
+ I2VA-related results are quite good. Ref2VA and T2V have not been tried yet.
64
+ Current tested weight range of approximately 0.9–1.2 works well.
65
+
66
+ The main materials are:
67
+
68
+ 1. Anal fisting
69
+ 2. Penis insertion (anal sex)
70
+ 3. Anal beads inserted into the anus
71
+ 4. Anal plug inserted into the anus
72
+ 5. Anal stretching / gape (self-spreading or being spread by another person)
73
+ 6. Anal creampie / semen flowing out, etc.
74
+
75
+ All material images were generated through a self-developed WebUI that integrates Stable Diffusion.
76
+ It automatically combines random character LoRAs + concept LoRAs + template prompts using Qwen3.8 Flash Next, then performs AI image review (only keeping images scored 80 or above), followed by final manual selection.
77
+
78
+ Captioning was also done through another self-developed WebUI, where Qwen3.8 Flash Next automatically generated the captions, which were then manually reviewed and confirmed.
79
+
80
+ Because various tests showed that it is very difficult to make the model correctly understand the positions, three trigger words were used to train the relative relationships.
81
+
82
+ * Female character is referred to as **zxqmei**
83
+ * Anus is referred to as **zxqanus**
84
+ * Vagina is referred to as **zxqvagina**
85
+
86
+ ===========================================================
87
+
88
+ 由於測試MiniMax H3似乎分不清楚肛門與陰道的位置,以及不理解肛門相關的NSFW行為。
89
+
90
+ (PS:經常把肛門與陰道搞錯邊,但大多是把肛門當成陰道)。
91
+
92
+ 因此自己訓練這個lora。 這是我第一個嘗試性Lora。
93
+
94
+ 圖片素材全部也是AI生成。總共使用了266張圖片素材,全部使用靜態圖。
95
+
96
+ 由stable diffusion :
97
+
98
+ illustrious / anima 動漫生成(90%)
99
+
100
+ 3D圖片生成(5%)
101
+
102
+ 真人照片(大概5%) 。
103
+
104
+ I2VA相關的部分,效果不錯,Ref2VA與T2v沒有嘗試過。
105
+
106
+ 權重目前測試約0.9~1.2效果良好。
107
+
108
+ 主要的素材都是
109
+
110
+ 1.肛門拳交
111
+
112
+ 2.陰莖插入肛交
113
+
114
+ 3.肛門珠插入肛門
115
+
116
+ 4.肛塞插入肛門
117
+
118
+ 5.肛門擴張(自行撐開或他人撐開)
119
+
120
+ 6.肛門射精流出等照片。
121
+
122
+ 全部素材圖由自行開發WebUI整合SD,由qwen3.8 flash next 自行隨機人物lora+概念lora+樣板Prompt自動生成素材,先AI審圖評分80分以上後再最後人工篩選。
123
+
124
+ 而標註部分也是由另一個自行開發的Webui,由qwen3.8 flash next自行標記,再最後人工確認。
125
+
126
+ 因為各種測試發現很難讓模型知道位置,因此用3個提示詞訓練相對關係。
127
+
128
+ 女性角色 用 **zxqmei 代稱**
129
+
130
+ 肛門 用 **zxqanus 代稱**
131
+
132
+ 陰道 用 **zxqvagina 代稱**
133
+
134
+ ---
135
+
136
+ Author: [acer78961204675](https://civitai.com/user/acer78961204675)
137
+
138
+ Model: [CivitAI Model Page](https://civitai.com/models/2974434?modelVersionId=3397856)
139
+
140
+ Archive: [CivArchive Page](https://civarchive.com/models/2974434?modelVersionId=3397856)
141
+
142
+ <!-- Version: 20261011_upload -->