声明：模型禁止任何形式的转载和二次发布，供学习参考经验交流，可自由融模，不得用于商用或其他非法用途。封面猫娘是莉喵，路过的大家都可以来ruarua她的头然后点个赞wwwThe cover cat girl is Leako, and everyone passing by can SQUISH THAT CAT and give her a thumbs up :3<ul><li>中文简介：</li></ul><h3 id="heading-6">（我愿称之为模型大杂烩，边用边融的产物，不小心出了个相对还不错的融模。）</h3>模型基底已经无从考究，我目前能记得的有Cetus、Orange、AnythingV5、PVC、Lyriel、LigneClaire等可能还有更多，权重数值几乎是脑内配比，少量测试，所以自认为这是依托答辩，但还挺好用的。该模型是一个偏向二次元画风的融模，能稳定的输出细节较多的二次元画风人物，整体画风如示意图所示，喜爱的朋友可以按需取用。依托于融了很多千奇百怪的模型，这个融模对tag的识别度还是相当好的，基本不是很怪的词都可以识别并在画面里体现，可以很轻松的达到你想要的效果，还有很多奇妙画风模型，关键词打对的时候还会有惊喜。并且，因为融了不少3D设计以及风景等现实模型，所以生成纯风景/物品相关图片时也能达到很精美的效果。推荐参数：正面tag推荐用masterpiece, best quality起手，反面常用low quality, worst quality等都可，可以参考例图的负面，屎山负面大欢迎，EasyNegative大欢迎。CFG我一般用默认7，往上往下走其实也都可以，大家可以试试有没有其妙效果。采样器测下来其实都挺适合的，各有所长吧，自用最多是Euler（出图快）、Euler a（出图快）、DPM++ 2M SDE Karras（质量好）、DDIM（手崩的少）。不用高清修复出图还可以，如果人物或者物体太小的话细节会崩，建议用下高修，个人目前一般常用Latent (bicubic antialiased)、R-ESRGAN 4x+ Anime6B，前者能出更多小细节但容易瞎长，后者比较好稳定高修之前的图片特征。v3b内置了一个840000的VAE，色彩比较浓厚，喜欢清淡的可以自行调整至其他。v3e2忘记放VAE了，使用时需要手动添加一下，推荐ClearVAE。<ul><li>English Introduction：</li></ul>(I would like to call it a hodgepodge of models, the product of merge while using, and accidentally producing a relatively good merged model.)The model base is no longer available, I can currently remember Cetus, Orange, AnythingV5, PVC, Lyriel, LigneClaire, etc. There may be more, the weight value is almost from my mind, a small number of tests, so I would call it a piece of ****, but it is quite useful.The model is a merged model biased towards the ACG painting style, which can stably output the ACG painting style characters with more details, and the overall painting style is shown in the pictures, and you guys can take it as needed.Relying on a lot of amazing models, this model is still quite good at the recognition of prompts, basically not very strange words can be recognized and reflected in the picture, can easily achieve the effect you want, there are many wonderful styles, will be surprises when the keywords are right. In addition, because many 3D designs and realistic models such as landscapes are merged in, it is possible to achieve beautiful results when generating pure landscape/object-related images.Recommended parameters:The positive prompts recommends using "masterpiece, best quality" to start, negative commonly used "low quality, worst quality," etc., or you can use negative prompts from sample images, and EasyNegative is welcome.CFG I generally use the default 7, up and down can actually be good, you can try to see if there is a wonderful effect.All the samplers are actually quite suitable, each has its own strengths, self-use is Euler (fast drawing), Euler a (fast drawing), DPM++ 2M SDE Karras (good quality), DDIM (less hand collapse).If the character or object is too small, the details will collapse, it is recommended to use the Hires.Fix, personally generally used Latent (bicubic antialiased), R-ESRGAN 4x + Anime6B, the former can produce more small details, the latter is better to stabilize the picture characteristics before the Hires.Fix.v3b version Built-in an 840000 VAE, the color is relatively strong, if you like light, you can adjust to others. And I forgot to add VAE in v3e2, need to set VAE manually when using, recommend ClearVAE.

BananaCatMix

bananacatmix_v4b3p.safetensors

v4b3p

v3e2

<a target="_blank" rel="noopener noreferrer nofollow" href="https://civitai.com/models/73820?modelVersionId=204126">https://civitai.com/models/73820?modelVersionId=204126</a>

Files with SHA256: 21de4b51ec29ac678eb234dd52d7ff78a16f39faa79dff4404e58f7e8e4f6b58