百花缭乱/midjourney 二次元/midjourney anime style Lora - v1.0

乱花渐欲迷人眼，浅草才能没马蹄。

——唐·白居易

大家好，我是chosen鲷。

这是我从p站上精选了上百张我很喜欢的midjourney二次元模型AI图进行训练的结果，不过倒不全是AI图，也夹杂了一些画师画的图，不过风格还是能够一致的。因为训练集足够的大，总体来说能出的东西还是蛮多的，从充满设计感的复杂构图人像到场景图再到一些特定角色，都可以的，不过猫猫狗狗确实画不出来，这次的训练集里没有，我后面会专门训练小动物的模型。

然后关于抽卡，我想有些人可能对我的模型产生了误解，我说我的模型可以抽卡玩，并不代表它只能抽卡玩，基本上对于各种tag还是有很不错的服从度的，这次的很多图都是写了tag的，跟各路lora相性也能蛮不错,所以你们可以看到例图里面有很多的碧蓝人物，可以抽卡意思是使用起来很简单，不需要写那种长篇大论的tag就可以出来很好的图，我是一直推崇简单化tag的，因为现在有 EasyNegative这样的强大的负面pt模型，其实大大的简化了质量和负面的书写。

建议的tag

, (masterpiece:1.2), best quality,midjourney,

这里写你想画的内容

<lora:midjourney_20230624181825:0.9>

Negative prompt: EasyNegative, badhandsv5-neg,

Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2301681042, Size: 512x768, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: ESRGAN_4x

建议的大模型，和我的chosen mix搭配比较适合出二次元风格的图偏向PIXIV上看到的那种，如果和ghost搭配会生成更有体积感的脸型，大家可以根据喜好使用，其他的像是irses impasto都行，不过跟三次元模型搭配效果一般，不是很搭，总体来说是个二次元模型。VAE挂载pastel，如果你用ghost，则不需要挂vae。

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https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0，https://www.pixiv.net/users/40911695 谢谢。

Hi, I'm chosen 鲷.

This is the result of training hundreds of my favorite midjourney quadratic meta-model AI diagrams from the p website, but not all of them are AI diagrams, and there are also some drawings drawn by artists, but the style is still consistent. Because the training set is large enough, generally speaking, there are quite a lot of things that can be produced, from complex composition portraits full of design sense to scene drawings and some specific roles, but it is true that cats and dogs cannot be drawn, which is not included in this training set. I will train small animal models later.

Then about the drawing card, I think some people may have a misunderstanding of my model, I said that my model can draw cards to play, it does not mean that it can only draw cards to play, basically for all kinds of tags or have a very good degree of compliance, this time a lot of drawings are written tag, It also works well with various LORAs, so you can see that there are a lot of blue characters in the example diagram. Drawing cards means that it is very simple to use, and you can make a good picture without writing a long piece of tag. I always advocate simplifying tag, because there are powerful negative pt models like EasyNegative. In fact, it greatly simplifies quality and negative writing.

Suggested tag

(masterpiece:1.2), best quality,midjourney,

Write what you want to draw here

0.9 > < lora: midjourney_20230624181825:

Negative prompt: EasyNegative, badhandsv5-neg,

Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2301681042, Size: 512x768, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: ESRGAN_4x

The proposed large model, with my chosen mix, is more suitable for a quadratic style of graph biased to the one seen on PIXIV, if it is matched with ghost, it will generate a more volumetric face shape, you can use it according to your preferences, others like irses impasto can be used, but with the three-dimensional model, the effect is general. It's not a good fit. It's a quadratic meta-model in general. VAE is mounted with pastel, if you use ghost, you do not need to mount vae.

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, https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0, https://www.pixiv.net/users/40911695, thank you.