Version Introduction

xifu V1.0: This version has the most extensive training dataset, resulting in a wider variety of generated outfits that are relatively accurate. However, its model applicability is relatively lower.

xifu V2.0: In this version, the training dataset was streamlined and regularization techniques were applied to improve generalization. Additionally, noise was introduced, resulting in more vibrant colors compared to V1.0.

xifu_reanime v2.0: This version builds upon xifu V2.0 by adjusting the weights of each layer to better suit the anime style, reducing the contamination of the artistic style. It is recommended to use AWpainting as the base template. Recommended weights range from 0.8 to 1.

Each version has its own advantages and disadvantages. Choose according to your needs.

Hand problems cannot be solved temporarily. You can add my negative embedding: badhand_xifu, which can be used in combination with other hand repair embeddings to avoid some pictures with large hand problems.

Model Introduction

Chinese Beijing Opera, an important form of traditional Chinese drama, is renowned for its unique music, singing, dancing, and theatrical performance style. The performances revolve around four main character types: Sheng (male roles), Dan (female roles), Jing (painted face roles), and Chou (clown roles).

Sheng roles typically portray male characters such as heroes, officials, and soldiers. They exhibit a masculine performance style, employing martial arts movements and body language to convey strength and power.

Dan roles, on the other hand, depict female characters such as noblewomen, consorts, and young girls. Dan roles emphasize delicate expressions and graceful dance movements to convey the gentleness and charm of women. Hua Dan is a specific type of Dan role that portrays beautiful and elegant female characters.

Jing roles encompass various characters, including eunuchs, villains, and knights-errant. Jing roles excel in humor, comedy, and dramatic movements, using quick and agile dance actions to express a wide range of emotions.

Chou roles are the comedic characters in Beijing Opera, often portraying ugly, funny, and comical roles. Chou performers bring joy and laughter to the audience through exaggerated actions, comedic behavior, and extravagant costumes.

Generation suggestions

2.5D and 3D Model Recommendation Weight: 0.5-0.7. For 2D models, you can choose a lower weight to add costume elements to the characters.

Trigger words:

Sheng Roles:

Basic trigger words: xifu, shengjue

Secondary trigger words (based on personal understanding, #explanations in parentheses):

wusheng #(martial role)

laosheng #(elderly role)

xiaosheng #(young male role)

hongsheng #(painted face role)

Dan Roles:

Basic trigger words: xifu, danjue

Secondary trigger words:

huadan #(young female role)

huashan #(flower dress)

wudan #(martial female role)

daomadan #(sword-wielding female role)

laodan #(elderly female role)

qingyi #(green-robed role)

Jing Roles:

Basic trigger words: xifu, jingjue

Chou Roles:

Basic trigger words: xifu,choujue

版本介绍：

xifu V1.0：该版本的训练数据集最广泛，生成的服装更多样，也较为准确，但模型适用性相对较差。

xifu V2.0：该版本训练时精简了数据集，采用了正则化，提高了泛用性，同时增加了噪声，画面色彩相对V1.0更鲜艳。

xifu_reanime v2.0: 该版本在xifu v2.0的基础上调节了各层权重，使其能较好的适用于二次元风格，减少了画风的污染。推荐使用AWpainting底模。推荐权重：0.8-1。

各版本各有优劣，按需选择，手部问题暂时无法解决，可以加上我的负面embeddding：badhand_xifu，与其他修手的embedding结合使用，可以避免部分手部问题较大的图片。

模型介绍：中国京剧是中国传统戏曲的重要形式，以其独特的音乐、唱腔、舞蹈和戏剧表演风格而著称。京剧的表演围绕着四个主要角色展开：生、旦、净和丑。 本模型包含了生、旦、净和丑四个角色（包括其下属分类）的服装。

生角通常扮演男性角色，如英雄、官员和士兵。生角的表演风格充满阳刚之气，善于运用武打动作和身体语言来展现角色的力量和气势。

旦角则扮演女性角色，包括贵妇、妃子和少女。旦角通过细腻的表情和柔美的舞姿展现女性的温婉和柔情。

花旦是旦角的一种特殊类型，扮演娇美的女性角色。 净角扮演多种角色，如宦官、恶棍和侠客。

净角的表演风格擅长幽默、滑稽和戏剧性的动作，以及快速敏捷的舞蹈动作来表达角色的喜怒哀乐。

丑角是京剧中的喜剧角色，通常扮演丑陋、滑稽和搞笑的角色。丑角通过夸张的动作、滑稽的行为和夸张的装束带给观众欢乐和喜剧效果。

2.5D和3D模型推荐权重：0.5~0.7，2D模型可以选择较低的权重来为人物添加戏服元素。

触发词：

生角：

基础触发词：xifu, shengjue

次级触发词（该分类基于个人理解，括号内为触发词的说明）：

wusheng（武生）

laosheng（老生）

xiaosheng（小生）

旦角：

基础触发词：xifu, danjue

次级触发词：

huadan（花旦）

huashan（花衫）

wudan（武旦）

daomadan（刀马旦）

laodan（老旦）

qingyi （青衣）

净角：

基础触发词： xifu, jingjue

丑角：

基础触发词：xifu, choujue