v1.0版本的简介见“About this version”

Introduction of v1.0 version in "About this version"

云墨丹心的触发词为：

Fu Hua(AZ), Chinese clothes,long sleeves,neck collar,belt,boots,hair over one eyes,hair ornament,hairpin,single earring, cloak,

其中的“cloak,“若是显现效果差、或是难以触发，则需要提高模型的权重，或提高”cloak,”自身的权重。

可以加入 “face,“来获得更贴近角色的脸部。

本次lora还训练了云墨的伞，有需要把“Oil-paper umbrella,“将入prompt即可。

推荐权重：0.75-0.9

Nsfw推荐权重：0.7-0.8

若脱衣不够干净可以尝试把“Chinese clothes,long sleeves,neck collar,belt,cloak“加入到negative prompt

换装推荐权重：0.6

和上面一样，若原衣服元素仍有残留，同样需要将“Chinese clothes,long sleeves,neck collar,belt,cloak“加入到negative prompt

此模型基于“animefull-final-pruned.ckpt”，不同大模型适配程度见图9

主要更新内容：

1、换用了更大分辨率的底图，现在的出图的细节效果会比第一版好不少

2、服装的局部可更换度更高，并有了新的内容加入，如披风和伞

3、专门训练了”Face,”来提高脸部的还原度

4、更换了训练的底模，现在在不同大模型下的泛用性会更好一点

For the " Azure Empyrea" prompt:

Fu Hua(AZ), Chinese clothes,long sleeves,neck collar,belt,boots,hair over one eyes,hair ornament,hairpin,single earring, cloak,

If the depiction of the cloak appears lackluster or proves challenging to generate, it may be necessary to increase the weight assigned to "cloak," or increase this lora weight.

Optional prompts:Face, Oil-paper umbrella,

Recommended weights:

- General: 0.75-0.9

- NSFW: 0.7-0.8

- For outfit changes: 0.6

- If residual elements of the original clothing persist, consider adding "Chinese clothes,long sleeves,neck collar, belt,cloak," in the negative prompt.

This model is based on the "animefull-final-pruned.ckpt",and the adaptation degree of different models can be seen in Figure 9.

Key updates include:

1. Adoption of higher-resolution base images resulting in significantly improved detail in the generated outputs compared to the initial version.

2. Enhanced capability for partial costume replacement, including the addition of a cloak and umbrella.

3. Specialized training for facial features to enhance facial restoration.

4. Application of a different base model for training, thereby improving the versatility across various base models.