Huggingcute | Anima

💡 Creator's Note & Model Story / 制作思路与说明

在使用 Anima 的期间，我一直想要复刻一种稳定且较为可爱的画风。但是奈何使用纯 LoRA 的话可能需要叠加 6-7 个画风 LoRA，如果再加上细节 LoRA、滑块 LoRA，最终生成一张图可能会需要用到 10-20 个 LoRA。而 LoRA 加载得越多，模型的语义理解能力就会下降很多。因此我在思考：如果将大模型进行融合来初步还原底层画风呢？经过一段时间的测试，效果确实比叠加 6-7 个画风 LoRA 更好。

While using Anima, I had been trying to recreate a stable and cute art style. However, relying purely on LoRAs meant stacking 6–7 style LoRAs. Add in detail LoRAs and slider LoRAs, and generating a single image could require 10–20 LoRAs in total. The more LoRAs loaded, the worse the model's text comprehension (prompt adherence) becomes. This made me wonder: what if I merged Checkpoints to recreate the underlying style first? After a period of testing, the results were indeed much better than stacking 6–7 style LoRAs.

🧩 Merged Checkpoints / 融合的大模型

• MiaoMiao Harem 1.5

• Adonis Anima 2.0

• Q-ANIMA 2.0

Q-ANIMA 是我最喜欢的 Anima 模型，它本身就拥有可爱的画风和稳定的肢体与构图。因此我选择了一个稳定且强大的模型以及一个男性模型来进行 Merge，稍微微调了一下 Q-ANIMA 的原始画风，但构图方面依然保持偏向 Q-ANIMA 本身。

Q-ANIMA is my favorite Anima model—it intrinsically features a cute style along with stable anatomy and composition. Therefore, I chose a powerful, stable model and a male-focused model to merge, subtly adjusting Q-ANIMA's original art style while preserving its excellent composition tendencies.

🎨 Integrated LoRAs / 融合的 LoRA

此外，在大模型 Merge 完成后，我也融入了 2-3 个画风 LoRA 来进一步稳固画风。除了我自己制作的 LoRA 外，还包含了其他创作者的优质 LoRA：

• Anima - Jock Studio / Zael / Mikkoukun | Style

Additionally, after merging the base models, I baked in 2–3 style LoRAs to further solidify the aesthetics. Along with my custom LoRAs, I integrated this fantastic LoRA by another creator:

• Anima - Jock Studio / Zael / Mikkoukun | Style

❤️ Credits & Acknowledgments / 致谢

综上，非常感谢这些大模型制作者和 LoRA 制作者！(●'◡'●)

All in all, huge thanks to all the amazing Checkpoint and LoRA creators mentioned above! (●'◡'●)