这个画风我真的超喜欢(´▽`ʃ♡)ƪ，试了试训练了一下，效果还挺不错的。虽然我用的素材不算多，训练的步数也一般，但自己觉得已经挺满意了(｡•̀ᴗ-)✧。我特意把素材集里的水印和签名都去掉了，所以生成出来的图特别干净(˶ᵔ ᵕ ᵔ˶)。建议大家提示词里用简单一点的背景，这样更有原版那味儿(◍•ᴗ•◍)。LoRA权重的话，控制在0.75到0.85之间效果最好(๑•̀ㅂ•́)و✧。不需要写触发词或者特殊提示词，直接加载就能用(•̀ᴗ•́)و。如果有侵权问题随时找我，我会立刻下架(｡•́︿•̀｡)。如果需要我代练lora可以联系我噢 咸鱼 余彧zyxI'm really in love with this art style (´▽`ʃ♡)ƪ. Gave it a try and trained a small model—turned out pretty nice actually. I didn't use that many training images or go for a huge number of steps, but honestly I'm already quite happy with the results (｡•̀ᴗ-)✧. I made sure to remove all watermarks and signatures from the dataset beforehand, so the outputs come out super clean (˶ᵔ ᵕ ᵔ˶). My tip would be to use simple backgrounds in your prompts—it keeps that original vibe intact (◍•ᴗ•◍). As for the LoRA weight, I found that 0.75 to 0.85 is the sweet spot (๑•̀ㅂ•́)و✧. No trigger words or special prompts needed—just load it up and it works (•̀ᴗ•́)و. If there's any copyright concern, feel free to reach out and I'll take it down right away (｡•́︿•̀｡).

Files with SHA256: 11e755cc397efa1e6583b7930bcf84d9414364bd007b385d95f57112ed7cbef1