基于《rimworld》社区mod内种族“ratkin”的同人图所训练的lora模型。新增了基于“Animagine XL V3.1”模型训练的sdxl lora，经测试效果良好，在其他模型上的效果我也不清楚。以下是sdxl版本的介绍。 <h2 id="heading-447">sdxl版本</h2>事实上Animagine XL V3.1本身就可以生成鼠鼠，然而ratkin标志性的米老鼠大耳朵却是力所不逮。当然或许可以生成，但我还是要炼一个lora。这个lora主要固定了ratkin的耳朵特征，尾巴不在我的考虑范围内——你可以选择加或不加尾巴。tag可以参考例图。所用到的其他模型有：<a target="_blank" rel="ugc" href="https://civarchive.com/models/216190?modelVersionId=247778">LCMTurboMix_DPM_SDE_Karras</a>、<a target="_blank" rel="ugc" href="https://tusi.cn/models/712847189698747537">Fixhands_anime_bdsqlsz_V1</a>、<a target="_blank" rel="ugc" href="https://civarchive.com/models/118418/negativexl">negativeXL_D</a>。你不一定要用这些模型，提到它们只是因为我用了。 <h3 id="heading-448">推荐的参数</h3>采样方法：DPM++ SDE Karras，因为我用了相关的lora。步数：8-16，这是相关lora决定的，如果图片质量不理想你可以自行调整。CFG：4，概念复杂时可以适当提高。以下例图提示你应该选择1.0-000016或者1.0版本。<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4c996e58-84ed-48ee-9bc3-4b237b9415b3/width=525/4c996e58-84ed-48ee-9bc3-4b237b9415b3.jpeg" /><h2 id="heading-449">sd1.5版本</h2>用了大约44张图片训练的模型，图片有经过我手动处理，但tag是自动生成的。本质上是我随便做的一个模型，质量我无法保证。ratkin是游戏《rimworld》的一个mod种族，有着迪士尼老鼠一般的大耳朵和尾巴。我觉得大耳朵很可爱因此制作了这个模型。注意，这并非某个特定的角色，我想要的是具有辨识度的大耳朵，因此你可以说这不是个角色模型。触发词：NewRatkinPlus强烈建议在negative prompt中添加：text, username, logo, multiple views, monochrome,尤其是monochrome，由于训练集中包含许多黑白图片，所以建议加上这个词以避免生成黑白图片。<h3 id="heading-450">Translated by Google:</h3>The lora model trained based on the fanart of the race "ratkin" in the "rimworld" community mod.A new sdxl lora based on the "Animagine XL V3.1" model training has been added. The test results are good, but I don't know the effect on other models.The following is an introduction to the sdxl version. <h2 id="heading-451">sdxl version</h2>In fact, Animagine XL V3.1 can generate mouse girl, but ratkin's iconic Mickey Mouse big ears are beyond its capabilities. Maybe it can generate, but still I want to train a lora.This lora mainly fixes the ratkin's ear features, the tail is not a consideration for me - you can choose to add a tail or not.Tag can refer to the example picture. Other models used are: <a target="_blank" rel="ugc" href="https://civarchive.com/models/216190?modelVersionId=247778">LCMTurboMix_DPM_SDE_Karras</a>, <a target="_blank" rel="ugc" href="https://tusi.cn/models/712847189698747537">Fixhands_anime_bdsqlsz_V1</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/118418/negativexl">negativeXL_D</a>. You don't have to use these models, I mention them just because I used.<h3 id="heading-452">Recommended parameters</h3>sampler: DPM++ SDE Karras, because I used the related lora.steps: 8-16, this is determined by the relevant lora. If the picture quality is bad, you can adjust it yourself.CFG: 4. It can be improved appropriately when the concept is complex. The following example shows that you should choose version 1.0-000016 or version 1.0.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/77562d27-42db-46a1-90dd-b397983bcc24/width=525/77562d27-42db-46a1-90dd-b397983bcc24.jpeg" /><h2 id="heading-453">sd1.5 version</h2>I used a model trained with about 44 pictures. The pictures have been manually processed by me, but the tags are automatically generated. It is essentially a model I made casually, and I cannot guarantee the quality.Ratkin is a mod race in the game "rimworld", with big ears and tail like a Disney mouse. I think big ears are cute so I made this model. Note that this is not a specific character, I wanted recognizable big ears, so you can say this is not a character model.Trigger word: NewRatkinPlusIt is strongly recommended to add in negative prompt: text, username, logo, multiple views, monochrome,Especially monochrome, since the training set contains many black and white pictures, it is recommended to add this word to avoid generating black and white pictures.

Files with SHA256: f03e354a518c126956757d53d59ae00d40f51797e2d3bad0ce1c47c9d4e70f06