大家好，我是chosen鲷，这次为大家带来的是モ誰画风的大模型。这其实是一个尝试性质的产物，因为这是我第一次尝试使用dreambooth去训练微调大模型，正好原来训练过モ誰的画风，那个模型由于也是我刚开始学习训练时候的产物，所以有诸多不足，例如手脚容易崩坏之类的，所以我正好借这个机会重新训练这个画风。モ誰老师是我非常喜爱的一位画师，本次训练所有训练集均来自于老师的p站图。如同例图所示，本模型非常适合画偶像的大师的角色（円香 浅仓透等），故打上了THE IDOLM@STER SHINY COLORS标签，但是它其实和其他动漫人物的相性也非常好，基本上任意人物都可以搭配这种脸型使用使用。<s>这个模型经过我的初步测试，画风还原度还不错，而且跟我各种lora搭配起来兼容性也还可以，所以我分享出来，由于我目前还在尝试更多的训练参数组合，所以后续可能还会推出V2版本，敬请期待。</s>V2相比于V1版本主要的改动有： 1.扩充了训练集（数量加倍了），似的画风还原度更高了，色彩也更鲜艳 2.经过修建让它回到了一个半精度模型应有的大小（2g）顺便一提，这个模型nsfw的表现也很不错。关键词是：who推荐的参数：Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 512*768, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+感谢大家的下载使用，如果喜欢的话欢迎关注我，我的bilibili pixiv ID均为chosen鲷，欢迎大家关注我的频道<a target="_blank" rel="ugc" href="https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0，https://www.pixiv.net/users/40911695">https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0，https://www.pixiv.net/users/40911695</a> 谢谢。The following content is from Google Translate：Hello, I'm chosen 鲷, and this time I'm bringing you a large model of モ誰 .This is actually an experimental product, because this is the first time that I tried to use dreambooth to train a large fine-tuning model, and it happened that I had trained モ who's painting style before. Since that model was also the product when I just started to learn and train, there were many shortcomings in it, such as hands and feet being easy to break and so on, so I just took this opportunity to retrain this painting style.モ誰 Teacher who is one of my favorite painters. All the training sets of this training are from teacher's P-station drawings. As shown in THE example figure, this model is very suitable for the role of the master of drawing idols (Hiroka Asakura Toru, etc.), so it is labeled with THE IDOLM@STER SHINY COLORS, but it actually has a very good resemblance to other anime characters, basically any character can be used with this face shape.After my preliminary test, this model is not bad, but also compatible with my various lora, so I share it, because I am still trying more training parameter combinations, so the follow-up may also launch V2 version, please look forward to it.By the way, this model nsfw also performs very well. The key word is: whoRecommended parameters:Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 512*768, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+Thank you for downloading and using, please follow me if you like, my bilibili pixiv ID is chosen Snapper, Welcome to focus on my channel, <a target="_blank" rel="ugc" href="https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0">https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0</a>, <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/40911695">https://www.pixiv.net/users/40911695</a>, thank you.

Files with SHA256: 3a61f5808bd1e78a288fabc25b4c880ddfa8d9b5a5cf321b6843538afc231e31