spouse NSFW
Details
| Model Type | Checkpoint |
| Base Model | SD 1.5 |
| Published | 9/3/2023 |
Download Files
About this version
Characters, Scenery, Realism, Artistic style
When I first came into contact with AI drawing, I had a dream. I wanted to draw a set of yoga action pictures. I did it, but it was still not perfect. In addition to many actions that were not drawn, it was too easy to have missing hands and feet. This is the case for everyone I have experienced it, so I don’t need to repeat it, so I applied the environment settings of the master to draw pictures, but unfortunately I still lacked hands and feet. Many actions were still not drawn, and because it was the original creation of the great god, I didn’t have it when I posted it. Let him show the picture information, but the yoga beauty is still quite beautiful, after all, she is the product of a master, haha
After trying to draw yoga and failed, I suddenly thought that although this website can see pictures and information of other people's works, but after all, it is a rigid text. If there is a function like a chat software, everyone can exchange information with each other, everyone can get into the situation quickly, like taking a helicopter from elementary school to high school, so they can quickly experience the fun of AI drawing.
So after thinking about it, I can only create a forum for everyone to communicate, which should be the best solution. Although there is no chat room function, the information provided by the picture information allows everyone to learn from each other freely
It's a pity that I have become a computer idiot more than 20 years ago, and now I am familiar with basic computer operations, but the problem is that I am not a graduate of a computer science department, and I can't write programs at all, so it is absolutely impossible to create checkpoints
And the computer I am currently using is a top configuration 3 years ago, but it still only has a 6G VRAM. The Internet guru explained that it is difficult to train Lora with this performance. Fortunately, it is difficult because I am not interested in training Lora at all. I just like doodles
In the end, I thought that the only way is to merge the models, so during this period of time, I tried to draw pictures with checkpoints everywhere. Personally, I prefer real-life models, so I only try for real-life models, and learn the drawing style and skills of that model by the way.
After continuous learning and experience in merging modules, I think this is my favorite. Of course, I also tried various painting styles, characters, landscapes, art, etc. After trying it, I felt that the ugly daughter-in-law can come out to see her in-laws, so I dared to take it out. otherwise, any master here who writes a program can tell the shortcomings of my module casually, so masters, I just love to doodle pictures, and I want to communicate with some players. Welcome to visit me, if you are Those who come to kicking my ass, please be merciful, don't say that my mod is worthless, but I will not respond to any of yours, haha
As I said before, I just want everyone to have more communication space, so I will not respond to any criticisms of my merged module or my approach. I said that I can’t write programs, so I have no way to respond. That's why idiots responded to you, here is very free, you can come if you like, and you can leave at any time if you don't like it
There is nothing here, just the word "freedom", you can draw whatever you want, good or bad, no one dislikes you, you can exaggerate as much as you want, but if someone reports it, I can only be forced to delete the picture, but don't worry, If the reported picture is more exaggerated or exceeds the scale, it’s okay, I will naturally not delete the picture, because the word "freedom" is emphasized here
As for those advertisers, please do whatever you want, as long as they don’t excessively destroy the page or harass the players, the principle of "freedom" is still the word. In the end, sponsors are exempt. As I said before, I can't even write any program , I can only draw ghosts, I can't cooperate with you, please do it yourself
Having said so much, I finally want to talk about the key points. The following is my experience after trying various painting styles. I hope it can help you get into the situation quickly and use this merge module faster and more easily.
There are two characteristics of this checkpoint that I like very much. One is that the face shape changes greatly by using only prompt words, and it is less limited by the few faces generated under the limited fixed frame. The other is that the vagina is not too bad when generated, and it is slightly It can be beautified by adding modification, see the following introduction will specifically mention this point
A piece of good news, when using sex toys, either use the "Euler a" sampler or LORA. I have tried many times, but I found that this checkpoint is in the "DPM++ 2M Karras" drawing without LORA. It has a high probability of appearing in pictures. It may be because the prompt words mentioned "sex toy" and other related prompt words many times. I will study this further. In short, I can easily generate pictures in the 2M sampler, which scares me a little.
1. For this checkpoint, I use 3 realistic checkpoints to merge, and one of them is a checkpoint for drawing Westerners, but the proportion is relatively low, so the characters born under unspecified circumstances will be mainly Orientals , but you can still vaguely see the outlines of Westerners, and you will find out if you observe more, but don't ask me which 3 checkpoints to merge and the ratio, I will not answer you, (because the original god must be respected, because You must respect the original master, because you must respect the original master), it is very important, so I say it three times
2. This checkpoint is integrated into VAE when merging. My computer runs at a very normal speed. I often wait for the time when the image is generated until I fall asleep, so I have fallen asleep especially in the past few months. Haha, for people who have not installed VAE, the speed will be slow Slower, so I’ll help you imagine it. What I integrated is the most commonly used "vae-ft-mse-840000-ema-pruned"
3. This is a realistic mod. If you want to increase the realistic effect, you can add "BadDream, (UnrealisticDream:1.2)" in the negative prompt, which will have a better effect. Using "surrealism" in the prompt will also have a good effect , the effect is better when using both at the same time, especially for landscape painting.
4. Please try to use high-quality prompt words. The prompt words with poor environment settings not only have poor image quality, ugly characters, but also small breasts, which need + weight to become larger, but if it is a high-quality prompt and negative prompt Under the words, "The key word is a good figure or a good body proportion", as long as two words are mentioned, there is no need for + weight or () symbols, and there is no need to mention the relevant prompt words of "breasts", size You don’t need to worry about it, AI will give birth to a pair of good sisters who are big and straight, please imagine for yourself, you can refer to my picture information, many of them hardly mention breasts, and give birth to a pair of good sisters that women envy, Don't ask me why, I already said I just merged and didn't know how to program. Maybe it's because of my character, haha
5. This checkpoint responds well to light, so there is no need to place too much emphasis on prompt words such as light source or lighting. In special circumstances, it is still necessary to strengthen the lighting. I found a problem in yesterday's test, that is white skin, don't use "high gloss and high lighting", I almost got blinded by the light, it is too bright, I can't see the characters clearly, this point should be paid attention to
6. The responses to prompt words and weights are also very friendly. When I was a novice, I wanted to do a lot of movements, but I was often troubled by weight adjustments. I also wrote a lot of prompt words, thinking that beautiful women and good figures would appear, but I ended up writing too many. Maybe the AI is confused, and monsters are often born. Now the skill has improved, but the weight is no longer concerned, but I still tested it and the response is good.
7. What everyone cares about the most is "beautiful", especially the face. Everyone wants to give birth to a supermodel girl. What a coincidence, it is easy to do. The keyword can be "(extremely sexy detailed and beautiful face:1.4)" Class prompt words, weight 1.3-1.4 can be enough, this is enough, no matter how much + weight and () symbol change, it will only affect the weight ratio
8. I have never used "controlnet", a good thing. I have to admit that it is a good thing and very practical. However, I am lazy and only used it a few times when I was a novice. I found it troublesome and stopped using it. So, how does this checkpoint react to controlnet? I'm really too lazy to try it, so please try it yourself. By the way, there is a very powerful function in it, which is used by many masters. It is the "adetailer". Do you want to know how to use it? By the way, I will give you a link, and watch the online teaching yourself, I come from the heart of Buddha, haha
9. As for my most commonly used high-definition upscaler "Hires upscaler: 8x_NMKD-Superscale_150000_G", I simply like the picture quality it draws. If you are interested, you can download it from the link below. Please take it yourself. There are many in the website High-definition zoom-in good things, if you are interested, you can also download other ones to use
https://huggingface.co/uwg/upscaler/tree/main/ESRGAN
After downloading, put it in the folder of the installation path, and then restart Stable Diffusion to enter the interface and it will appear automatically
"C:\novelai-webui-aki-XXX\models\ESRGAN"
10. When I first started playing Stable Diffusion, I was crazy. When I saw the painting style I liked, I downloaded checkpoints and LORA desperately, and my computer was quickly overwhelmed. Now I don’t pursue these things, so I rarely use LORA in recent works. It is only used in specific situations, and the most commonly used one is lora:add\_detail:1, this kind of plug-in, but I have tried a few, and the response is not bad, but if you want to say how good the effect is, please try it yourself
11. Finally, and most importantly, apart from wanting a beautiful face, I believe everyone is very concerned about the other one, which is the "pussy". Most people should be no exception, after all, "food and sex are also", In the online world where no one can see you, the nature of the sex devil is exposed, but unfortunately, drawing the vagina is much more difficult than the face, and I often want to vomit when I draw it, so everyone often sees my works. It's not that I have this habit of exposing pussy, but I won't deny it when I say I'm evil. Haha, it's actually to check whether the environment setting of the cue words is good or not. Good cue words can indeed give birth to a beautiful pussy, but that also requires If you are skilled, even if I have used the prompt words of the masters, there will still be times when the drawing fails, so here is a shortcut for everyone. Although it is not perfect, it can greatly improve the picture quality. The prompt words can be used like "beautiful detailed and realistic pussy skin texture" to improve the picture, this is a prompt I came up with after a long time of personal research, it is simple and clear, but it cannot guarantee perfection, please use it at your own discretion
12. People will ask me why I use the name "Spouse". It's very simple, because I have no wife, no girlfriend, and I have always been single. I only have the computer to accompany me for more than 20 years after work every day. The computer is my lover, my wife, easy enough my girlfriend
Finally, I would like to say that if the English translation in this article involves words such as racial discrimination, please forgive me. After all, I don’t know the taboos of each race or country. Simply using the wrong words does not mean that I have racial discrimination. Discriminatory view
剛接觸AI繪圖時,有個夢想,想繪製一套瑜珈動作套圖,我做到了,但是還是很不完美,除了很多動作沒畫出來以外,就是太容易出現缺手缺腳畫面,這情形大家都經歷過,就不用我贅述,所以我套用大神的環境設定去畫圖,但還是很遺憾的缺手缺腳,很多動作仍是沒畫出來,也因為是大神的原創,所以當初我發帖時沒讓他顯示圖片資訊,不過瑜珈美女還是挺美的,畢竟是大神的產物,哈哈
在嘗試畫瑜珈失敗後,我又突發奇想,這網站雖能看到別人作品的圖片資訊,但畢竟是死板板的文字,要是能有像聊天軟體的功能,大家在裡面互相交流資訊,是不是大家都能很快進入狀況,類似搭直升機般從小學升到高中生,這樣能快點體驗AI繪圖的樂趣
所以我思來想去只有創建1個版面提供大家交流,應該是最佳的解決方案,雖然沒聊天室功能,但圖片資訊提供的訊息,可以讓大家自由的互相學習
可惜我從20幾年前的電腦白癡,到現在對電腦基本操作都駕輕就熟,可問題我畢竟不是學電腦的科系畢業,完全根本不會寫程式,所以創作檢查點那是絕對不可能了
而我目前用的電腦是3年前算頂配的,可也還是只有6G顯示卡,網路大神解說,說這效能訓練LORA有困難,幸好有困難,因為我對訓練LORA完全沒興趣,我就是只喜歡亂畫圖
最後我想到唯一的方法就是合併模型的方式,於是這段時間我到處嘗試檢查點來畫圖,個人偏好真人模組,所以只針對真人模組去嘗試,順便學習那個模組的繪圖風格跟技巧
在不斷的學習跟合併模組經驗下,覺得這個我最喜歡,當然我也嘗試各種畫風,人物,風景,藝術等等都嘗試過後,覺得醜媳婦可以出來見公婆了,才敢拿出來現醜,不然這裡隨便一個大神寫個程式,都能隨便說出我這模組的缺點,所以大神們,我只是愛亂畫圖,並且想多跟一些玩家交流,大神來光臨我歡迎,如果是來踢館的,請手下留情,別把我這模組說的一文不值,但我不會回應你的任何動作,哈哈
前面說了,我只是想讓大家多個交流空間,所以對我這合併模組的任何批評,還有我的作法,我都一概不予回應,都說了不會寫程式,根本無從回應,所以傻子才回應你們,這裡很自由,喜歡來就來,不喜歡隨時可以離開
這裡什麼沒有,就是"自由"2個字,你想畫什麼都可以,好壞都沒人嫌棄你,多誇張都隨你,不過要是有人檢舉,我也只能被迫刪圖,但是放心,如果被檢舉的圖,別人有更誇張或是超過尺度的都沒事,我自然也不會去刪圖,因為這裡講求"自由"2個字
至於那些廣告商請隨便,只要不過分破壞版面或騷擾玩家,還是以"自由"2個字為原則,最後,贊助商就免了,我前面都說了,我半個屁程式都寫不出,就只會鬼畫符,我無從跟你合作起,請自便
說了這麼多,終於要說重點了,以下是我嘗試各種畫風後的心得,希望能幫助大家快速進入狀況,更快駕輕就熟的使用這個合併模組
這檢查點有2個特性我很喜歡,一是只用提示詞臉形變化大,比較不會被侷限固定框架下生成的那幾張臉限制,另外一個就是陰部生成時並不會太差,稍加修飾便可美化,看以下介紹會特別說到這點
插播一個福音,使用性玩具時,不是使用"Euler a"採樣器,就是要用LORA,我嘗試過很多次了,但我發現這檢查點在"DPM++ 2M Karras"繪圖又沒+LORA情形下,高機率出圖都會出現,有可能是提示詞中多次提到"sex toy"等相關提示詞原因,這我會再研究,總之能在2M採樣器輕易生成圖片,這點讓我有點驚嚇
這個檢查點我採用3個寫實畫風的檢查點進行合併,其中有1個是畫西方人的檢查點,不過比例較低,所以在不指定情形下生出來的人物會以東方人為主,但還是會隱約看到西方人的輪廓,多觀察你就會發現的,但是別問我用哪3個檢查點合併還有比例,我不會回答你,（因為要尊重原創的大神,因為要尊重原創的大神,因為要尊重原創的大神）,很重要所以說3次
這個檢查點在合併時融入了VAE,我的電腦跑起來速度很普通,經常等圖生成的時間等到睡著,所以最近幾個月特別會睡,哈哈,對於沒有安裝VAE的人速度會更慢,所以幫你們設想好了,我融入的是最常用的"vae-ft-mse-840000-ema-pruned"
這是寫實模組, 如果你想增加寫實效果可以在負提示加入"BadDream, (UnrealisticDream:1.2)",會有更佳效果,在正提示中使用"超現實主義"畫風也有不錯效果,2者同時使用效果更好,尤其是畫風景類
提示詞請盡量高質量,較差的環境設定的提示詞不止畫質較差,人物醜,而且胸部變的很小,需要+權重才能變大,但是如果是高質量的正負提示詞下,"關鍵字是好身材或是良好身體比例",只要提到2個就夠了,也不需要+權重或是()符號,也不需要提到"breasts"的相關提示詞,大小也不需要管,AI自己會生出一對又大又挺的好姐妹,請自己想像,可以參考我的圖片資訊,很多幾乎都沒提到胸部,就生出一對令女人都羨慕的好姐妹,別問我為什麼,都說了我只是合併,不會寫程式,或許是我人品好吧,哈哈
這檢查點對光線反應不錯,所以無須太強調光源或是照明之類提示詞,在特殊環境下還是會需要加強照明,昨天測試發現一個問題,就是白皮膚,不要在+高光澤以及高照明,害我差點被亮瞎眼睛,太亮了,都看不清楚人物,這點要注意
對提示詞跟權重的反應也滿友善的, 以前新手時想要做很多動作,經常被權重調整所困擾,還寫一堆提示詞,以為會出現美女跟好身材,結果寫太多,可能AI都困惑了,生出怪物也是經常發生,現在功力變好了,反而權重已經不在意,但還是測試了,反應不錯
大家最在意的就是"beautiful", 尤其是臉,大家都想生出超模正妹,真巧,很容易就辦到,關鍵字可以用"(extremely sexy detailed and beautiful face:1.4)"這類提示詞,權重1.3-1.4都可,這樣就夠了,+再多權重跟()符號變化不會太大,只會影響權重比例
我從來不用controlnet這好物,不得不承認這是好物,非常實用,但是我懶,只在新手時用過幾次,嫌它麻煩就不再用了,所以這檢查點對controlnet反應如何,我真的懶得試,請自己嘗試,順便介紹一下裡面有個很強大功能,很多大神在用,就是"adetailer" 這好物,想知道怎用嗎？順便給你們連結,自己看網路教學,我佛心來的,哈哈
- 至於我現在最常用高清放大"Hires upscaler: 8x_NMKD-Superscale_150000_G",只是我單純喜歡它畫出來的畫質,如果你們感興趣,可到下面的連結網站下載,請自行服用,網址內有很多高清放大好物,有興趣的也可以下載其他的來用
https://huggingface.co/uwg/upscaler/tree/main/ESRGAN
下載完後放到安裝路徑的資料夾內,然後重啟一次Stable Diffusion進入介面就自動出現
C:\novelai-webui-aki-XXX\models\ESRGAN
剛開始玩Stable Diffusion時很瘋狂,看到喜歡的畫風,檢查點跟LORA都拼命下載,很快就塞爆我電腦,現在反而不追求這些,所以我近來作品都很少用LORA,只在特定情況才使用,最常用的反而是lora:add\_detail:1,這類插件,但還是嘗試過幾個,反應不算差,但要說效果有多好,還是請您自己嘗試
最後,也是最重要的,大家除了想要1張美人臉以外,相信都很關心另外1個,就是"陰部",應該絕大部分人都不例外吧,畢竟"食色性也",在沒人看的到你的網路世界,色魔本性都暴露無遺,但很遺憾,畫陰部比臉難太多了,經常畫出來的都想嘔吐,所以大家常看到我的作品都是開腿露陰部,並非我有這癖好,但是說我色我不會否認,哈哈,其實是要檢查用的提示詞環境設定好壞,好的提示詞確實能生出很美的陰部,但那也是要有功力的,即使我套用過大神的提示詞,也同樣會有畫失敗的時候,所以這裡提供大家一個捷徑,雖不完美,但能大幅改善畫質,就是提示詞可以用類似"beautiful detailed and realistic pussy skin texture"來改善畫面,這是我個人研究很久後才想出來的提示詞,簡單明瞭,但是無法保證完美,請自行斟酌服用
大家會問我為何用"Spouse"這個名稱,很簡單,因為我沒老婆,沒女朋友,一直單身一人,每天下班就只有電腦陪伴我20幾年,電腦就是我的愛人,我老婆,我女朋友,夠簡單吧
最後我要說,文中的英文翻譯如果有牽涉到種族歧視等類用詞,請原諒我,畢竟我不知道每個種族或是國家的禁忌,單純就只是用詞錯誤,並不代表我有種族歧視的眼光
Model description
新年快乐,终于赶上新年发布改良版,这次改良在兼顾写实风格下加入Q版元素,增加画面了丰富性,整体效果还不错,这版本（v2.1）跟上一版本(v1.0) 的差别如下：
**1.提示词输入"(nsfw)",你想看到的通常都会显灵,该裸露的几乎不费力都能如愿,除非是人物穿太多衣服或是使用过多LORA,删除衣服并加上想要的部位即可,所以最近我把这模型改为"NSFW",因为太容易达阵了
2.个人偏好不爱用"ADetailer"以及"ControlNet",这2个功能很棒,只是我懒,完全不用,有时LORA也懒得用,所以只要点"(Copy Generation Data)",完全套用我的图片设定即可,除非特定情形,不然都会分享提示词,另外与v1.0版一样,完全不用管胸部大小,提示词也不用,包君满意,可以多参考我的帖子
3.写实模型不容易出现的画面效果及道具,现在都很容易出现,另外画风景图片时,有更梦幻的效果
4.人物可能没有v1.0 版画的美,但变的比较可爱,Q版
5.画面整体变丰富了,不论人物眼神、表情、手势、姿势、身体彩绘、衣服、道具或是背景画面都有明显效果
6.四肢及手指也有明显改善,当然,仍然还是有瑕疵,还请大家多包涵
7.人物的身材也更美了,比较少出现婴儿肥,容易出现细腰身材
**
Happy Chinese New Year, finally released an improved version in time for the New Year. This improvement adds Q version elements while taking into account the realistic style, which increases the richness of the picture. The overall effect is not bad. This version (v2.1) is the same as the previous version (v1.0 ) are as follows:
1. Enter the prompt word "(nsfw)", what you want to see will usually appear, and what you want to be naked can be done with almost no effort, unless the character wears too many clothes or uses too much LORA, delete the clothes and add what you want The parts are enough, so I recently changed this model to "NSFW" because it is too easy to score a touchdown.
2. Personal preference is not to use "ADetailer" and "ControlNet". These two functions are great, but I am too lazy to use them at all. Sometimes I am too lazy to use LORA, so just click "(Copy Generation Data)" and fully apply my image settings. Unless there are specific circumstances, the prompt words will be shared. In addition, just like the v1.0 version, there is no need to worry about the size of the breasts and the prompt words are not used. I guarantee that you are satisfied. You can refer to my posts more.
3. Picture effects and props that are not easy to appear in realistic models are now easy to appear. In addition, when drawing landscape pictures, it has a more dreamy effect.
4. The characters may not be as beautiful as the v1.0 prints, but they have become cuter, Q version
5. The overall picture has become richer, no matter the eyes, expressions, gestures, postures, body paintings, clothes, props or background pictures of the characters have obvious effects.
6. The limbs and fingers have also improved significantly. Of course, there are still flaws, so please bear with me.
7. The figure of the character is also more beautiful, with less baby fat and prone to slim waist.