Misakura Rin (御樱禀) | Sakura no Toki (樱之刻) | Pony
Details
Download Files
Model description
这是一个以Pony为底模训练的樱之刻画风的御樱禀的Lora，可以召唤还算OK的大禀老师，不含其他角色。
This is a LoRA trained with a Pony base model in the Sakura no Toki art style, focusing on Misakura Rin. It can summon a reasonably accurate version of Rin-sensei. No other characters are included.
Noob XL的版本拥有更加优秀的表现，见suggested resources。
Noob XL version has much better performance, see suggested resources.
————————————
Tags
Pony底模常用(Commonly used with the Pony base model)：
score_9,score_8_up,score_7_up,score_6_up,score_5_up,score_4_up
角色(Character)：
Misakura_Rin, Sakura_no_Toki_art, red hair ribbon
在效果不好时推荐搭配(If results are poor, try combining with)：
1girl, solo, brown eyes, brown hair, medium breasts
黑色礼服（唯一一套樱之刻中的服饰）(Black dress (the only outfit from Sakura no Toki))：
Misakura_Rin_Outfit, Sakura_no_Toki_art, white pantyhose
在效果不好时推荐搭配(If results are poor, try adding)：
black_dress, long sleeves
选用后大概率还原原作表情(To better match original expressions)：
closed eyes, closed mouth, open eyes, open mouth
————————————
推荐底模(Recommended Base Models)：
Hassaku XL (Pony)
Mikoshi [Pony]
————————————
注意！Pony底模相关：
需要使用SDXL VAE。
本人使用的VAE下载于PonyV6的附件：
Pony Diffusion V6 XL
/model/257749/pony-diffusion-v6-xl
需要将CLIP终止层数设置为2。
Important! Regarding Pony-based models:
Requires SDXL VAE.
The VAE I use is from the Pony V6 attachments:
Pony Diffusion V6 XL
/model/257749/pony-diffusion-v6-xl
Set CLIP Skip to 2.
————————————
推荐参数(Recommended Parameters)：
采样方法(Sampler)：Eular a
Steps：36 (24~40 is fine)
Width：1024 (768~1440 is fine)
Hight：1024 (768~1440 is fine)
Distilled CFG Scale：0
CFG Scale：7 (3~8 is fine)
————————————
推荐使用高分辨率修复(Hires. fix) (Recommended Use)：
放大算法 (Upscaler)：R-ESRGAN 4x+ Anime6B
Steps：15
重绘幅度 (Denoising Strength)：0.3~0.4
Hires Distilled CFG Scale：0
Hires CFG Scale：3 (1~5 is fine)
————————————
目前的版本显然不够完美：
嘴部表情难以还原导致作为人物最高辨识度的脸部五官容易不像；
Lora训练过拟合以及训练集不够多样导致有时手部动作单一，难以控制；
底模训练集导致的在涩涩的时候会出现黑色线条审核。
也许未来会更新。
Current Limitations:
Mouth expressions are hard to replicate, making facial features — the most recognizable part of the character — less accurate.
Overfitting in LoRA training and lack of variety in the dataset may lead to repetitive hand poses that are difficult to control.
Due to the training set of the base model, black censor lines may appear in NSFW images.
————————————
案例：
