[EN]This specialized version is released separately to avoid confusion with the regular version.
[CN]特化版单独发布以避免与原版混淆。
Recommended Settings(推荐设置):
Sampler(采样器)：Euler a
Quality tags: Not required
不需要正面的质量提示词。
I sincerely apologize for the delayed release of this new LoRA, which I completed training on July 17th...I simply couldn't find the time to publish it until now.
The Anima base model excels at precisely reproducing all the effects from the original reference images provided by 魔族娘 - 萨利希尔, including the Mozudoll body proportions. However, the Mozudoll figure is not the most moe-friendly design: its distinct secondary sexual characteristics restrict the full expression of moe aesthetics, and these established body features in previous LoRAs proved difficult to reverse through prompts alone. So I tried to carry out a personalized modification in the digital world.After image selection, LoRA retraining, Photoshop compositing, and liquify adjustments, I finally compiled enough "semi-realistic" artificial assets to train this moe-specialized style. Although it's still not chibi, the body proportions are more gender-neutral/nonbinary (though not perfectly achieved due to the base model and my own skill limitations), aiming to realize a moe aesthetic that transcends traditional gender frameworks.
This represents an artistic style the original Mozudoll never possessed. It carries latent potential that never had the chance to flourish, as its body dimensions were digitally locked by local technocratic mandates, even though the Mozudoll brand already conformed to these specific measurements even before the digital constraints were imposed.
After eight failed training iterations plagued by issues like overfitting and unintended 2.5D stylization, I've finally achieved a version that looks "authentic" enough.
魔族风格（ANI）萌系特化版
非常抱歉，我在7月17号炼好的新lora，因为挤不出时间，一直拖到现在才在这里发布。
Anima底模能更精确的还原魔族娘-萨利希尔提供的原素材图片的各种效果，而这当然包括了mozudoll的身材比例。但是mozudoll体不是最萌的那种，性征影响了萌元素的发挥，而通过提示词难以扭转在之前的lora里已经定型的体型。于是我尝试在数字世界里进行个人化的改造。在筛选图片、lora重炼、PS拼接与液化的处理后，终于有足够的“半真半假”的人工素材来炼出这种萌系特化的风格了。虽然仍不算Q版，但身材更性别中立（尽管由于底模和个人能力而不能完美实现），以便实现超越性别框架的萌系美学。
这是mozudoll本身不具有，有潜力有，但是没机会有的艺术风格，因为被当地技术官僚指令的身材数字锁死了——虽然mozudoll品牌本身在那道数字枷锁降临之前就已然契合该数字。
经历八个版本的失败训练（过拟合、2.5D化等）之后，我终于得到了这个“看起来”足够“真实”的版本。
Resist the systemic aesthetic bullying driven by lobbying politics via mechanisms of policy capture.
拒绝游说政治通过俘获机制造成的系统性审美霸凌。
Description
I sincerely apologize for the delayed release of this new LoRA, which I completed training on July 17th...I simply couldn't find the time to publish it until now.
The Anima base model excels at precisely reproducing all the effects from the original reference images provided by 魔族娘 - 萨利希尔, including the Mozudoll body proportions. However, the Mozudoll figure is not the most moe-friendly design: its distinct secondary sexual characteristics restrict the full expression of moe aesthetics, and these established body features in previous LoRAs proved difficult to reverse through prompts alone.So I tried to carry out a personalized modification in the digital world.After image selection, LoRA retraining, Photoshop compositing, and liquify adjustments, I finally compiled enough "semi-realistic" artificial assets to train this moe-specialized style. Although it's still not chibi, the body proportions are more gender-neutral/nonbinary (though not perfectly achieved due to the base model and my own skill limitations), aiming to realize a moe aesthetic that transcends traditional gender frameworks.
This represents an artistic style the original Mozudoll never possessed.It carries latent potential that never had the chance to flourish, as its body dimensions were digitally locked by local technocratic mandates, even though the Mozudoll brand already conformed to these specific measurements even before the digital constraints were imposed.
After eight failed training iterations plagued by issues like overfitting and unintended 2.5D stylization, I've finally achieved a version that looks "authentic" enough.
非常抱歉，我在7月17号炼好的新lora，因为挤不出时间，一直拖到现在才在这里发布。
Anima底模能更精确的还原魔族娘-萨利希尔提供的原素材图片的各种效果，而这当然包括了mozudoll的身材比例。但是mozudoll体不是最萌的那种，性征影响了萌元素的发挥，而通过提示词难以扭转在之前的lora里已经定型的体型。于是我尝试在数字世界里进行个人化的改造。在筛选图片、lora重炼、PS拼接与液化的处理后，终于有足够的“半真半假”的人工素材来炼出这种萌系特化的风格了。虽然仍不算Q版，但身材更性别中立（尽管由于底模和个人能力而不能完美实现），以便实现超越性别框架的萌系美学。
这是mozudoll本身不具有，有潜力有，但是没机会有的艺术风格，因为被当地技术官僚指令的身材数字锁死了——虽然mozudoll品牌本身在那道数字枷锁降临之前就已然契合该数字。
经历八个版本的失败训练（过拟合、2.5D化等）之后，我终于得到了这个“看起来”足够“真实”的版本。