我最喜欢的画师之一，可惜的是至今没有他的风格 LoRA，于是自己尝试做了一个。Takano Otohiko (aka icenotes) is one of my favourite artists. Unfortunately, there is no existing LoRA of his style. So I tried to create one myself.<h1 id="heading-3741">V0</h1>失败作，没能很好地还原风格。最明显的差别是生成的画面细节过多，难以重现画师许多原创插画中那种童话般的氛围。但其实生成的图片很多观感还不错，于是还是打算发出来，并命名为 V0。This attempt failed to replicate the style well. The most noticeable difference is the excessive refined details in the AI generated images, which often result in the loss of the fairytale-like atmosphere.Nevertheless, many of the generated images don't look that bad aesthetically. So I still tend to publish it and give it the name "V0".<h2 id="heading-3742">Recommendations</h2>模型以 NAI 为底模训练而来，应该在目前流行的绝大多数二次元模型上都有相近的表现。其余的参数可以参考样图中的信息。我还没有对这些参数进行微调测试，不过感觉效果还不错。此外，使用 "claris" 提示词，可以生成组合 Claris 的形象。但这并不是一个专门训练的角色模型，加上 SD1.5 生成二人图片的固有问题，所以效果并不稳定。NAI is used as the base model, so this LoHA should perform somewhat consistently in most popular Anime models nowadays. The rest of the my parameters can be found in the metadata of the uploaded sample images. I haven't tried to tweak most of them yet, but they did reasonably well from my experience.You can also try the prompt "claris", which can generate the Anime style characters of the duo Claris. But this is not intended to be a character LoRA, and SD 1.5 has some inherited problems with 2-person images. So the results are not stable.

Files with SHA256: 98b2140ab864957b0c07557ebd076df0dee9daf408f5742f4950ee540536b2fd