<h1 id="dop3hypersd1-di3r58ubk">Dop3HyperSD1</h1><h2 id="dop3theme-v2-merged-with-some-hypersd1.5-models.-77xt4osmw"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e367330c-80be-4a5b-bdca-1cd36b663ac0/width=525/e367330c-80be-4a5b-bdca-1cd36b663ac0.jpeg" />Dop3Theme v2 merged with some HyperSD1.5 models.</h2><h3 id="settings:-q4ngnnb15">Settings:</h3><ul><li>Prompt: A simple sentence or the typical style of prompting.</li><li>Negative: (low quality, ugly, simple, blurry:1.2), add more words when needed</li><li>Sampler: DPM++ SDE / Euler A / <s>DPM++ 3M SDE (some issue)</s></li><li>Schedule type: Uniform / SGM Uniform</li><li>Steps: 8 (4, 8, 12, usually works best)</li><li>CFG Scale: 2 (1-4)</li><li>Size: 512x768 or 768x512 (512x512, 512x768, 768x512)</li><li>Denoising strength: 0.4 (0.3-0.5)</li><li>ClipSkip: 1 &amp; 2</li><li>Hires upscaler: <a target="_blank" rel="ugc" href="https://civarchive.com/models/318385/4xrealisticrescaler100000g">4xRealisticrescaler_100000G</a></li><li>LoRA: <a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/Hyper-SD/blob/main/Hyper-SD15-8steps-CFG-lora.safetensors">HyperSD15-8step-CFG</a> STR: 0.2-0.34/8/12-Step LoRA = all works</li></ul>Negative Embedding:only when needed<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/embed/EasyNegative/tree/main">EasyNegative</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/AsciiP/badhandv4/tree/main">badhandv4</a></li></ul>I hope you enjoy this merger!<h3 id="last-notes:-9mj011lx6">Last Notes:</h3>Please, do not sell this checkpoint.Give me a shout out if you use this checkpoint in your own merges

Dop3HyperSD1

ClearVAE_V2.3_fp16.pt

dop3hypersd1_v10.safetensors

v1.0

This is a very versatile 1.5 mbw model merge. My preferred sampler is 2M Karras and i usually use clip skip 2. Play with different negative embeddings as it might change the style significantly. This model can generate anime 2D, 2.5D and 3D, depending on the prompt. I use the Stability vae vae-ft-mse-840000-ema-pruned.safetensors, but you can use whatever you like, but a vae is needed since i didn't merge a vae into the model. IT can make sfw and soft nsfw (meaning only nudity and not sex scenes).IMPORTANT : This model has add detailer 0.5 and noise offset 0.5 baked into it, so maybe don't use it ? It's untested anyway to use those loras again.Group of negative embeddings i used : bad-picture-chill-75v, image by bad-artist-anime, render By bad artist -neg, negative_hand-neg, verybadimagenegative_v1.3, easynegative, dangermouse, dangerdonkey, ng_deepnegative_v1_75t, (deformed iris, deformed pupils:1.1), (poorly drawn eyes:1.2), (poorly drawn iris:1.2), (poorly drawn pupils:1.2)You can use all of them, use you own, or mix however you like. Negatives change the style a lot.PS : Remember if you want to use the RArt model, use it with highres fix or upscale at 0.35 denoise with ReyMix or another good colors or details model. The reason is that the model, at least for me, gives a yellowish color. The issue is fixed completely with the suggestion above.Suggested prompt structure for both models :(Style) (in artstyle by artist X), (other type of style for example surrealism if you wish). BREAK(Composition) (A full body shot of a X girl for example). BREAK(Girl) description of the girl, BREAK(background) (description of background or just aesthetic and simple backgournd). BREAK(detail) (masterpiece, high quality), &lt;lora:RussianDoll:0.25&gt;

ReyMix

RArt

&lt;h3 id=&#34;heading-601&#34;&gt;&lt;strong&gt;&lt;em&gt;我创建了一个lora和发图交流的qq群欢迎大家加入：817686141&lt;/em&gt;&lt;/strong&gt;&lt;/h3&gt;&lt;p&gt;&lt;strong&gt;&lt;em&gt;&lt;span style=&#34;color:rgb(230, 73, 128)&#34;&gt;求求大家的返图和星星&lt;/span&gt;&lt;/em&gt;&lt;/strong&gt;&lt;/p&gt;&lt;h3 id=&#34;heading-3&#34;&gt;&lt;strong&gt;&lt;span style=&#34;color:rgb(250, 82, 82)&#34;&gt;如果使用不来lora推荐直接把我的原图丢进去识别&lt;/span&gt;&lt;/strong&gt;&lt;/h3&gt;&lt;h3 id=&#34;heading-602&#34;&gt;&lt;strong&gt;&lt;em&gt;这是风格一种艺术风格lora我推荐和夜枭大佬的xxmix系列使用或者是有奇幻风格的大模型。&lt;/em&gt;&lt;/strong&gt;&lt;/h3&gt;&lt;h3 id=&#34;heading-602&#34;&gt;&lt;strong&gt;&lt;em&gt;最好是配合conterlnet使用&lt;/em&gt;&lt;/strong&gt;&lt;/h3&gt;&lt;h3 id=&#34;heading-602&#34;&gt;&lt;strong&gt;&lt;em&gt;我的颜色风格系列之一，如你所见他的艺术风格具有很好的观赏性质，训练集有200张左右，总共用了12000步训练，lora内容十分风丰富推荐使用我例图的tag为基础进行改进，脸部我推荐用&lt;span style=&#34;color:rgb(250, 82, 82)&#34;&gt;After Detailer&lt;/span&gt;&lt;span style=&#34;color:rgb(243, 244, 246)&#34;&gt;插件，能生成更好的脸部。主要是mj素材练成，本身不带脸部数据，最好加上手部动作的描述，不然容易摸脸动作。权重给0.5-0.9，,会得到更好的效果，请关注我后续我会发上更多模型。如果你喜欢我的模型系列的话。帮我点个赞，喜欢我的lora的话可以给我发电支持我一下哦，你的支持是我前进的动力。&lt;/span&gt;&lt;/em&gt;&lt;/strong&gt;&lt;/h3&gt;&lt;p&gt;&lt;br /&gt;&lt;/p&gt;

Files with SHA256: 42a404c8851ce41db8d0c79d4476d17dfe9834e442f1640e9bbfb96bc57e1bb6