Importat remider: It is recommeded that all users who previously used
this LoRA dowload the ew V2.5 versio. The ew versio has bee adjusted
by LoRA Block Weight, which ot oly preserves facial effects but also reduces
cotamiatio of clothig, backgroud, ad style.
重要提醒：建議之前所有使用這個LoRA地用戶下載新地V2.5版本， 新地版本經過LoRA Block
Weight調整，既保留臉部效果又减小了對服飾、背景、風格地侵染。
The ame "Pure Iocet Girl " is give because it is ot a specific
character's face, but rather a bled of styles that ca produce a youg ad
beautiful iocet girl's face, ad a more atural ad realistic (less plastic
surgery sesatio) oval face; It is ecessary to eable Face restoratio ,
with a weight betwee0.25 ad 0.85. The higher the weight, the more likely
it is to produce childish features that look too childish. It is recommeded
to use it aloe with a value of 0.5, ad the maximum value should ot exceed
0.9. If abormalities occur whe usig it with other models, Lora, or
certai samplers, it is recommeded to radomly adjust the weight.Otherwise,
there will be deviatios that affect the style. Suitable for semi realistic
ad realistic effects, but ot tested i aime styles.
If you like this LoRA, please leave a commet or image!
取名叫“純真女孩”
，是因為它不是特定人物的臉，而是更像一種風格混合，能產生年輕美麗的清純少女臉，並且是更自然真實（整形手術感覺更少）的鵝蛋臉；一定要開啟臉部修復，權重在0.25～0.85之間
，越高越容易產生看起來太孩子氣的幼稚特徵，單獨使用建議0.5 ，最高不要超過0.9
，否則會產生偏差影響風格。如果搭配其它模型、Lora或者某些採樣器使用發生了异常，建議隨機調整權重。適合半寫實和逼真效果，但並沒有在動漫風格中測試。
** 如果你喜歡這個LoRA，請留下評論或圖片！**
O May 23rd,
Xidi V2.5 versioKow problem: "Log Rage Full Body" will collapse o the
lower face.
5月23日，
新地V2.5版本已知問題：“遠景全身”下臉部會崩。
O April 22d,
The V2 versio is the fial versio of this LoRA, with the followig
adjustmets made:
The model was traied with "stablediffusio 15 prued " ad "vae ft
mse-840000_ " to improve the uiversality of LoRA;
Trim tags, remove some tags ad add ew tags such as clothig ad posture,
while simplifyig the abbreviatio keyword to "PI_girl ". Of course, as
log as LoRA is loaded or the etry is writte, ca be used without triggerig words;
Updated the dataset, removed some potetially harmful images, added some high-
quality high-defiitio ew images, traied with about 300 images , ad
also improved the traiig resolutio;
Icrease the dim value to 64, icrease the umber of traiig steps, ad
adjust the learig rate.
Kow issue: Usig excessive weights (exceedig 0.7) ca still affect the
character's clothig ad backgroud style.
Example of dataset labels:
1girl，best quality，masterpiece，realistic，solo，lookig at
viewer，face，eyes，ose，mouth，lips，smile，makeup，
Pose: suggestive spread_ legs_ pose，lyig，o
back，Stadig，sittig，keelig，_ Squattig，
Clothig: Ice pricess dress,
4月22日，
V2版本是這個LoRA的最終版本 ，做了以下調整：
模型改用“** stablediffusio15-prued_** ”與“** _vae-ft-mse-840000_**
”訓練，以提高LoRA的泛用性；
修剪標籤，删除了一些標籤和新增了一些服飾、姿勢等新標籤，同時簡化縮寫關鍵字為“PI_girl
”，當然，只要加載了LoRA或者詞條寫上，無需觸發詞也可使用；
更新數据集，删除了一些可能導致不良的影像，添加了一些高品質的高清新影像，用了約300張圖片 訓練，同時也提高了訓練分辯率；
增大dim值調整為64，新增訓練步數，以及調整了學習率。
已知問題：使用過高權重（超過0.7）依然會影響人物服裝和背景風格。
數据集標籤示例：
1girl，best quality，masterpiece，realistic，solo，lookig at
viewer，face，eyes，ose，mouth，lips，smile，makeup，
姿勢： suggestive spread_ legs_ pose，lyig，o
back，Stadig，sittig，keelig，Squattig，_
服飾： Ice pricess dress，
O April 10th,
The V1 versio is a experimetal LORA model that used approximately 200
images to trai i "ChiloutMix" ad geerate "vae-ft-mse-840000" VAE. This is
a very featureless LoRA, which did ot meet my origial expectatios. It still
looks like a mixed race with may Korea faces.
4月10日，
V1版本是一個實驗性的LORA模型 ，用了約200張圖片 在“** ChilloutMix** ”中訓練和“ vae-ft-
mse-840000 ”VAE生成。 這是一個非常沒有特點的LoRA，並不符合我原本的預期，看起來還是混合了很多韓國面孔的混血。
评论