如果你喜歡我的工作可以幫我買杯咖啡，謝謝
If you like what I do, you can buy me a cup of coffee. Thanks.
私の仕事を気に入っていただけたら、コーヒーを一杯おごってください。 ありがとうございます。
https://ko-fi.com/vicentma
Why do this?
Basic theory: Realistic model is better for human body structure, animation model is better for movement and more artistic, I hope to get the advantages of both.
Use the realistic base model as the overall base model, use LORA to adjust the format of the drawing from 3D-PHOTO--->2D-DRAW.
Adjust LORA parameters to find the sweet point.
V2.0
After understanding the LECO training method, we will adjust the parameters and the learning rate in order to achieve better results with smaller files and fewer weights.
The available weights are from +2 to -2, and the recommended weights are from +2 to -1.0, with positive values providing animation tendency and negative values providing real tendency.
The performance of positive weight (animation) is better than negative weight (real).
比較了解LECO訓練法了之後將調整參數和學習率希望用較小的檔案和較少的權重達到更好的效果
可用權重在+2~-2，建議權重為+2~-1.0，正值提供動漫畫的傾向，副值提供真實的傾向
在正權重（動畫）的表現比較負權重（真實）好
LECO学習法を理解した後、より小さなファイルとより少ない重みでより良い結果を得るために、パラメータと学習率を調整する。
利用可能な重みは+2～-2、推奨される重みは+2～-1.0で、正の値はアニメーション傾向、負の値は現実傾向を示す。
正のウェイト（アニメーション）は負のウェイト（現実）よりも良い結果をもたらします。
V1.0
It's best to use it on 2.5D models, with a weighting of -7~7, which is large than usual . Negative values will decrease the realism, positive values will increase it.
If you want to use it for realistic models or purely anime models, please adjust the weight range.
If you want to use it for real models or pure anime models, please adjust the weighting range.
The example is calculated using unstableinkdream 7.5 balanced, an average 2.5D model.
A LYCORIS trained in the LECO method for transferring from photo 3D style to graphic animation style.
Training based on this git
https://github.com/p1atdev/LECO
最好在2.5D模型上使用,權重-7~7，對你沒看錯很大，負值會降低真實度，正值則是增加
如果你想要用在真實系模型或者純動漫系模型,請調整權重的區間
範例是利用unstableinkdream 7.5 balanced 這個平均型2.5D模型算出來的
https://civitai.com/models/1540?modelVersionId=40626
一個利用LECO方式訓練出來的從照片3D風格移轉到繪畫動畫風格的LYCORIS
依據此工具進行訓練
https://github.com/p1atdev/LECO
