⭐If you have the means, please do support the original artist's creative work!
⭐如果您有条件，请支持原创艺术家的创作！

A LoRA that reproduces the drawing style of the artist M-Da S-Tarou (MだSたろう)
一个目的是还原画师M-Da S-Tarou (MだSたろう)绘画风格的LoRA。

Artist page: https://www.pixiv.net/users/3316400

Tips

- Strength 1.0 is the full value, not a midpoint — do not exceed it (the magnitude is extrapolated and faces/hands break). Use 0.85–1.0.
~Using moss image SFT for inference seems to yield better results~

I personally use the SFT version for T2I and I2I

- No trigger word. Just describe the image.

- Trained on portrait crops (≈64%)

⚠WHIT 4 STEP TURBO LORA⚠:
Recommended sampling: cfg 1.5, 8-12 steps.

At cfg 1.0 the source artwork's light mosaic gets reproduced; use censor tag in negative prompt and cfg 1.5, it will disappears.

About floating hearts

The artist's floating hearts are context-driven, not baked in: a neutral expression with white/simple background drops them entirely, or use Negative prompt.

小提示

- 强度 1.0 就是满值，别超（超了幅度外推，脸和手会崩）。建议 0.85–1.0。

~用moss image SFT进行推理效果似乎会更好一些~

我本人T2I、I2I用的就是SFT版本
- 无触发词，直接描述画面。

- 训练用竖构图较多（≈64%）

加载4步TUROB LORA时：
推荐采样：cfg 1.5 / 12 步。 cfg 1.0 时原作的薄码会被复现，负面提示词中加入审查词并将cfg 1.5 时完全消失；

关于爱心特效：

画师的漂浮爱心是**情境触发**的，不是焊死的——中性表情 + 纯白底/简单背景就会完全消失。或者直接将其加入负面提示词。