生成時請參考圖片、觸發詞及版本說明欄

更新了針對單一服裝訓練的較穩定版本

版本號為V2前綴

之前下載過多套衣服版本、覺得不夠穩定的用戶請再次下載

非常抱歉，由於我對於多服裝同個LoRA的訓練仍缺乏經驗，

目前仍在摸索中……

若造成困擾，先向各位致歉，

如有良好的訓練控制方法，懇請不吝賜教。

Updated a more stable version for single outfit training

, the version number is V2 prefix

Please download the version of multiple sets of clothes before, and if you feel that it is not stable enough, download it again

I'm sorry because I still don't have much experience in training for the same LoRa with multiple clothes

It's currently being explored...

If it causes trouble, I apologize to you.

If you have a good way to control training, please don't hesitate to teach me

"我推的神明來了!" w

不愧是神裝，真的是神明級別的難弄

（也因為我正在研究如何將多件衣服放入同一個LoRA）

我在本地搭建了kohya，並針對不同衣服使用不同資料夾、對應不同步數與觸發詞來處理

但若這次把所有衣服都丟進去訓練、各自分步數訓練，總步數可能會高達上萬（=_=）

目前的折衷方案大概是拆分為多個LoRA

拆成兩個LoRA：一個是素材最多的兩件——校服與神裝，

另一個則是其他三件常服。

因為是第一次嘗試，不穩定的地方肯定不少，

尤其是神裝的提示詞，我多次與ChatGPT討論並持續修正，

我發誓我已經盡力了...orz

（其他已知問題請見版本說明欄）

但仍無法保證每次都能穩定出圖，請多多包涵。

非常期待看到大家用這個LoRA創作，歡迎踊跃上傳分享！

When generating, please refer to the reference image, trigger word, and imprint column

 "The god I pushed is come on!" w

It's worthy of a god costume, it's really a god-level difficulty

(ALSO BECAUSE ON THE ONE HAND IS STUDYING HOW TO MAKE MULTIPLE PIECES OF CLOTHING INTO A LORA RELATIONSHIP)

Build kohya locally, and use the number of out-of-sync folders and trigger words for different clothes to deal with

But like this time, if you throw all the clothes into the training, and train separately for different steps, the total number of steps may come to tens of thousands (=_=)

THE CURRENT COMPROMISE IS PROBABLY TO make multiple LORAs of diffrent clothes

Split two LORA, one is the two school uniforms and god costumes with the most train 
materials

and another one is the other three regular clothes

Because the first time I try, there will be a lot of places that feel unstable.

In particular, the prompt words of the god costume part have been asked many times with chatgpt and constantly corrected the prompt words.

I swear I've done my best... orz

(For other known issues, see the release notes section.)

But I still can't make a stable image everytime, please bear with me

I'd love to see you create with this LoRa, so please upload and share it!

聲明：

模型皆基於此篇文章的教學訓練而成，再次感謝！

純粹出於對喜愛角色的臨摹，為非官方創作，請勿用於任何商業用途。

如涉及任何版權問題，請通知我，我將立即刪除。

此LoRA雖未禁止NSFW內容，但使用本LoRA創作的相關責任請由使用者自行承擔。

Statement:
The models are trained through the education of this article, thank you again!

Purely based on the imitation of my favorite characters, it is not an official creation ,only fiction, and should not be used for any commercial purposes

If there is any copyright infringement problem, please let us know that it will be removed immediately

Although NSFW is not prohibited by this LORA, you are responsible for the part of your creation that uses this LORA