BoobAI

所有模型均基于NoobAi的Epsilon-Prediction（NoobE）。

・boobAI α_v2.0

在还原特定画风方面，α1.0更胜一筹。α2.0因追求真实感，导致还原特定画风的能力有所下降，这是一种权衡关系。

α1.0在再现特定画风上更优秀；相比之下，α2.0在追求写实表现的过程中，削弱了复制特定风格的能力。这是典型的权衡关系。

・boobAI v4.0

此模型具有真实且光泽的质感。

瑕疵较少，是v2.1的进化版本。

・boobAI α_v1.0

降低了真实感，更贴近原始noobAI的风格，增强了漫画质感，使画风更易调整。

降低了写实度，使风格更接近原始NoobAI，并增强了漫画质感，以便更容易调整画风。

若追求真实感，推荐使用v4.0；若追求独特画风或更强烈的漫画感，建议使用α1.0。

---

推荐参数：

采样器：euler a
调度器：Karras
步数：20 至 30
CFG Scale：5 至 7

EasyReforge (Reforge + dmd2_sdxl_4ste)
lora:dmd2_sdxl_4step:1.0
采样器：LCM
调度器：Karras
步数：8
CFG Scale：1

---

NoobIEater_v3.0 ＋ T-illunai3_v2 融合
base_alpha=0
Weight_values=0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
model α

NoobIEater_v3.0 ＋ T-illunai3_v2 融合
base_alpha=0
Weight_values=1,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,1,1,1
model β

model α + model β
base_alpha=0
Weight_values=0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5
boobai