Mochizuki Kei art style(望月KEI画风) | anima
Diffus
Generate with 70,000+ AI models
Run Stable Diffusion in your browser. No local setup, no GPU tuning.
Start creating↗
No-code AI image workflows
Prototype segmented image generation workflows, then turn them into APIs.
Build workflows↗
AI video ads from URLs or images
Create product and marketing videos with Veo, Sora, Kling and more in one place.
Make videos↗
Model description
Mochizuki Kei art style的画风lork
————————————
触发词（Trigger tag）
@Mochizuki Kei,————————————
推荐权重（Recommended weight）：0~1.3 (try 1 first)
目前测试了 factor=16 的版本——虽然模型体积只有 6.4MB，但整体表达能力感觉偏弱。如果你更看重模型体积、对画风还原度要求不高，可以先试试这个版本；后续我也会再训练一版 factor=10 的模型做对比
I've tested the factor=16 version — while the model is only 6.4MB in size, its expressive capacity feels somewhat limited. If file size matters more to you than strong style fidelity, this version is worth trying for now. I'll also be training a factor=10 version later for comparison
以下是对比图可以做个参考(角色Lora权重为1)
Here's a comparison image for reference (character lora weight set to 1)
四小只怎么变成三小只了
训练时为了尽可能还原真红的眼睛细节，CG部分的重复次数设成了2，这也导致角色lora本身包含的画风特征已经强到足以盖过lork画风本身的强度了，罪魁祸首还是因为F社出HD不会重新绘制立绘，导致立绘眼睛部分像素偏低、细节偏少(很难想象CG与立绘是同一个角色)