这可能是Stable Diffusion里最会讲故事，懂得情绪、气氛的SD大模型，GhostMix。

起初并未想给GhostMix单独做一期评测，只是随便找个模型试一试Lora（Add_detail）的效果。

但让人没想到的是，GhostMix给了我一批水准极高，情绪和故事感极佳的作品。于是，棕猫先生又陷入了2天10000张图的爆肝模式。

先说结论：

首先，GhostMix是Stable Diffusion最会讲故事，对情绪把控最好的模型之一。

其次，这个模型的作者是中国人，更应该支持。

最后，这款模型的复现率（抄作业）极高，参数准确的情况下，几乎可以达到100%还原。

本文图片、Lora较多，提示词放在图片下面

配置总结放在文章末尾

(shot from below:0.9),(cinematic look:1.55),night, [[soft cinematic light, adobe lightroom, photolab, hdr, intricate, highly detailed]], Wasteland, complex background, (depth of field), epic realistic, mystical haze, (by alan kenny, agnes cecile:1.2), 1girl, close-up,<lora:add_detail:1> Negative prompt: verybadimagenegative_v1.2-6400,asian,poorly drawn,bad anatomy,wrong anatomy,extra limb,missing limb,disconnected limbs,mutation,mutated,ugly,disgusting,blurry,cropped,text,(worst quality, low quality:1.4),bad anatomy,(watermark, signature, text, logo, watermark:1.3), Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 3567452634, Size: 448×832, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.3, ENSD: 31337, Version: v1.2.1, Hires upscale: 2, Hires steps: 20, Hires upscaler: 4x-UltraSharp

被埋没的【超级插画】大师模型

GhostMix并不是一个新的模型，从2023年3月11日开始，作者已经对该模型进行了多次更新。就目前测试来看，该模型在插画领域的优势十分突出。

特别是在色彩和与机甲领域，GhostMix可以说是创造了属于自己的独特风格。

1mechanical girl,((ultra realistic details)), portrait, global illumination, shadows, octane render, 8k, ultra sharp,metal,intricate, ornaments detailed, cold colors, egypician detail, highly intricate details, realistic light, trending on cgsociety, glowing eyes, facing camera, neon details, machanical limbs,blood vessels connected to tubes,mechanical vertebra attaching to back,mechanical cervial attaching to neck,sitting,wires and cables connecting to head, A mystical forest glowswith an otherworldly light, filled with towering, crystalline trees and delicate, glowing flowers. The air is filled with the sound of ethereal music and the fluttering of wings, as a myriad of magical creatures flit about the scene. A massive, shimmering portal hangs in the sky, leading to another realm filled with wonder and adventure. The entire scene is rendered ina dreamlike, impressionisticstyle, with vibrant colors and bold brushstrokes that capture the essence ofthe enchanted realm,<lora:add_detail:1> Negative prompt: NSFW,3d, cartoon, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normalquality, jpeg artifacts, signature, watermark, username, blurry, artistname, young, loli, elf, 3d, illustration ng_deepnegative_v1_75t Steps:30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 2511330957, Size: 512x768, Model hash: 9c321174ae, Model: ghostmix_v11, Denoising strength: 0.5, ENSD: 31337, Version: v1.2.1, Hires upscale: 2, Hires steps: 20, Hires upscaler: 4x-UltraSharp

从图片中可以看到，GhoistMix天生对色彩，质感，细节都有很强的基础素质。

再加上的神级Lora：add_detail的辅助，将细节又提升了大约20%，就形成了现在大家看到的细节拉满的图片。

同时，这款模型的良图率很高，据作者自己介绍，这款模型是在自己更新迭代了47次后分享出来的，良图率达到了50%以上。

根据棕猫先生测试，这款模型在近景的出图率有90%以上，中景约50%，远景约20%，算是很好的成绩了。可以当做测试提示词的主力模型使用。

当然，这款模型受到棕猫先生强力推荐的重要原因，还是因为他的氛围与故事感。

顶级的氛围感

这两组作品，棕猫先生并没有给出太详细的提示词，更多是让GhostMix自由发挥。

让棕猫先生惊讶的是，GhostMix对提示词：Wasteland|荒野，mystical haze|神秘的朦胧感理解是极为富有创造性的。

另外，这种创造性有10%的功劳归功于mocha2，一款专门画云彩和丁达尔光线的lora。

1girl,Girl in Red,Cloud Layer,China Tower,(shot from below:0.9),(cinematic look:1.55),night, [[soft cinematic light, adobe lightroom, photolab, hdr, intricate, highly detailed]], Wasteland, complex background, (depth of field), epic realistic, mystical haze, (by alan kenny, agnes cecile:1.2), <lora:add_detail:1>,<lora:loong:0.4> Negative prompt: verybadimagenegative_v1.2-6400,asian,poorly drawn,bad anatomy,wrong anatomy,extra limb,missing limb,disconnected limbs,mutation,mutated,ugly,disgusting,blurry,cropped,text,(worst quality, low quality:1.4),bad anatomy,(watermark, signature, text, logo, watermark:1.3), Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 3383181154, Size: 448×832, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Version: v1.2.1, Hires upscale: 2, Hires steps: 20, Hires upscaler: Latent 1girl, The background is a Chinese dragon in the clouds,(shot from below:0.9),(cinematic look:1.55),night, [[soft cinematic light, adobe lightroom, photolab, hdr, intricate, highly detailed]], Wasteland, complex background, (depth of field), epic realistic, mystical haze, (by alan kenny, agnes cecile:1.2), <lora:add_detail:1>,<lora:loong:1> Negative prompt: verybadimagenegative_v1.2-6400,asian,poorly drawn,bad anatomy,wrong anatomy,extra limb,missing limb,disconnected limbs,mutation,mutated,ugly,disgusting,blurry,cropped,text,(worst quality, low quality:1.4),bad anatomy,(watermark, signature, text, logo, watermark:1.3), Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 1599556351, Size: 448×832, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Version: v1.2.1, Hires upscale: 2, Hires steps: 20, Hires upscaler: Latent

在绘制龙的过程中，出图率不到1%。

原因在于Ai对于中国龙的理解太少了，它甚至分不出头与尾。loong这个Lora还不太成熟，大家看看就行。

1girl,Girl in Red,Cloud Layer,China Tower,(cinematic look:1.55),sunlight, [[soft cinematic light, adobe lightroom, photolab, hdr, intricate, highly detailed]], Seaside, complex background, (depth of field), epic realistic, The mysterious haze lifts, (by alan kenny, agnes cecile:1.2), <lora:add_detail:1>, <lora:mocha2:0.3> Negative prompt: verybadimagenegative_v1.2-6400,asian,poorly drawn,bad anatomy,wrong anatomy,extra limb,missing limb,disconnected limbs,mutation,mutated,ugly,disgusting,blurry,cropped,text,(worst quality, low quality:1.4),bad anatomy,(watermark, signature, text, logo, watermark:1.3), Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 2043490290, Size: 448×832, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.6, Clip skip: 2, Hires upscale: 2, Hires steps: 15, Hires upscaler: Latent, Lora hashes: “add_detail: 7c6bad76eb54, mocha2: c704614f2cdd”, Version: v1.3.1 1girl,Girl in Red,Cloud Layer,China Tower,(shot from below:0.9),(cinematic look:1.55),night, [[soft cinematic light, adobe lightroom, photolab, hdr, intricate, highly detailed]], Wasteland, complex background, (depth of field), epic realistic, mystical haze, (by alan kenny, agnes cecile:1.2), <lora:add_detail:1>,<lora:loong:0.4> Negative prompt: verybadimagenegative_v1.2-6400,asian,poorly drawn,bad anatomy,wrong anatomy,extra limb,missing limb,disconnected limbs,mutation,mutated,ugly,disgusting,blurry,cropped,text,(worst quality, low quality:1.4),bad anatomy,(watermark, signature, text, logo, watermark:1.3), Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 641371857, Size: 448×832, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Version: v1.2.1, Hires upscale: 2, Hires steps: 20, Hires upscaler: Latent

这张图片用了loong的Lora，但龙的效果很差，被棕猫先生用PS Beta处理了。

顺便说一句，PS Beta目前效果仅相当于SD去年9月的水平，用来生成和创作效果一般。但是用来处理瑕疵，那绝对是超一流的。

1girl,Girl in Red,Cloud Layer,China Tower,(cinematic look:1.55),sunlight, [[soft cinematic light, adobe lightroom, photolab, hdr, intricate, highly detailed]], Seaside, complex background, (depth of field), epic realistic, The mysterious haze lifts, (by alan kenny, agnes cecile:1.2), <lora:add_detail:1>, <lora:mocha2:0.3> Negative prompt: verybadimagenegative_v1.2-6400,asian,poorly drawn,bad anatomy,wrong anatomy,extra limb,missing limb,disconnected limbs,mutation,mutated,ugly,disgusting,blurry,cropped,text,(worst quality, low quality:1.4),bad anatomy,(watermark, signature, text, logo, watermark:1.3), Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 2043490290, Size: 448×832, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.6, Clip skip: 2, Hires upscale: 2, Hires steps: 15, Hires upscaler: Latent, Lora hashes: “add_detail: 7c6bad76eb54, mocha2: c704614f2cdd”, Version: v1.3.1

关于GhostMix的实用干货：

国产的GhostMix应该是最会讲故事的大模型，值得骄傲。

Add_detail是最值得装的lora，它对画面细节的提升较大。

除此之外，关于GhostMix有一些重要的配置，建议大家保存。

Vae推荐：kl-f8-trinart

Lora推荐：add_detail，参数1

提示词推荐：(masterpiece, top quality, best quality, official art, beautiful and aesthetic:1.2), (1girl:1.3), (fractal art:1.3),

色彩提示词推荐：(flat color:1.3),(colorful:1.3),(masterpiece:1.2), best quality, masterpiece, original, extremely detailed wallpaper, looking at viewer,1girl,solo,floating colorful water

高清修复：一定要做高清修复，建议2倍，重绘幅度0.4-0.5

采样器建议：DPM++ series； Steps: 20-30；CFG:5-7

分辨率建议：512×768，机械少女主题对分辨率设置非常敏感，不建议设太低的宽高比

最后，感谢大家的支持。

创作不易，如果您看到这里了，还请留下你的赞与关注。

模型与Lora获取老规矩，评论区见！