当前位置:首页|资讯|Stable Diffusion|Civitai

Stable Diffusion 常用模型的总结

作者:Hxic丨stable发布时间:2024-03-09

有话在先:我没有统计过哪些是不商用的,爷爷们自己找。

夸克链接:https://pan.quark.cn/s/9ef37802aa9c     里面是这个页面的txt文件




https://civitai.com/models/269232/aam-xl-anime-mix

  AAMXL

  动漫

  这基本上源自 AAM AnyLoRA Anime Mix,但它基于 SDXL。它不适用于 SD1.5 的 loras 和嵌入,即使它们在原始 AAM 或 AnyLoRA 上进行了训练。“混合”作为动漫风格的混合。我建议你使用 CFG 5-7(不高于 8),20-30 步和 Euler a。

  ComfyUI工作流程:https://pastebin.com/BHCEzc6T

  ##########################################################################

anything-v5-PrtRE

##########################################################################

AWPainting_v1.3

  动漫

  可使用GPT写自然语言的Prompts使用

  本模型适用性较广,插画、二次元及普遍的2D、2.5D均可很好出图,对niji风的复现,搭配LoRA nijiexpress以及不同LoRA的复用有很好的相性

  推荐搭配LoRA: Nijiexpress V1 V2,Xiaorenshu小人书,墨心moxin,沁彩

  推荐 Prompts:Fisheyes, masterpiece, top quality, best quality, official art, beautiful and aesthetic, animation

      Neg_Prompts: EasyNegativeV2, ng_deepnegative_v1_75t, (low_quality:1.4), (worst_quality:1.4), (badhandv4:1.1),collage, artist_name, signature, artist_logo, watermark

  ##########################################################################

https://civitai.com/models/44219/camelliamix25d

CamelliaMix_2.5D_V2  

  真人,女

  使用 vae : vae-ft-mse-840000-ema-pruned.safetensors, orangemix.vae.pt

  推荐设置

  采样方式:DPM++2M Karras, DPM++ SDE Karras, DPM++ 3M SDE Karras

  Clip skip : 2

  Hires steps : 13

  Hires.fix upscaler : 4x-UltreaSharp 

  CFG Scale : 6~11

  Prompt : [please refer to exmple images]

  Negative : (worst quality, low quality:1.3)

  ##########################################################################

https://civitai.com/models/15003/cyberrealistic

cyberrealistic_v42

  真人,女,男,年龄,房屋,动物,森林

  SDXL 版本

  该模型的主要优势之一在于它能够有效地处理文本反转和

LORA,从而提供准确和详细的输出。此外,该模型需要最少的提示,使其非常用户友好和易于访问。

  ##########################################################################

https://civitai.com/models/7152/facebombmix

facebombmix_v1Bakedvae

  动漫,2.5d,人物上半身,肖像

  从动漫到 2.5D 风格的封面。适合一般用途。这种 2.5D 风格最适合上半身的肖像(upper body)

  如果颜色褪色,请尝试使用VAE: https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt

  Positive : (masterpiece, sidelighting, finely detailed beautiful eyes: 1.2), masterpiece*portrait, realistic, 3d face, lustrous skin,

  Negative : (worst quality, low quality:1.4), watermark, logo,

  Model : FaceBombMix-fp16-no-ema.safetensors

  SD VAE : kl-f8-anime2.ckpt

  Clip skip : 2

  Sampling method : DPM++ SDE Karras

  Sampling steps : 32 (24 ~ 32, as you like)

  Upscaler : R-ESRGAN 4x+ Anime6B

  Hires steps : 14 (in my case, 7 ~ 14 weren't really different for R-ESRGAN 4x+ Anime6B)

  Denoising strength : 0.5 (I change around 0.45 ~ 0.55)

  Upscale by : 2

  CFG Scale : 9 (7 ~ 9, as you like)

  ##########################################################################

https://civitai.com/models/24383/grapefruit-hentai-model

grapefruitHentaiModel_grapefruitv41(逆天)

  色情,动漫

  Don't use face restore and underscores _, type red eyes and not red_eyes.(不要使用面部恢复和下划线 _,键入红眼而不是red_eyes。)

  If you don't want a open, spread or gaping pussy, it helps to avoid the word "pussy" in your prompts (that applies also on vaginal insertions). Prompt "nsfw" can push some things into the right direction and can spice some images up (motion lines, floating hearts, steam, but adds sometimes censoring).

  每个构建为在 anyV3 或 orangeMix 上运行的 LoRA 也适用于葡萄柚。

  Black result fix (vae bug in web ui): Use --no-half-vae in your command line arguments

  ##########################################################################

https://civitai.com/models/34469/indigo-furry-mix

indigoFurryMix_v115Anime

  雄性龙/毛茸茸的内容/男性焦点/NSFW/半色情,可控制

  male dragons/furries content/Male focus/NSFW

  V90 artist styles here:https://mega.nz/folder/NxJBWKpS#LZFVyOyvgFIAj2307-GDkQ

  V105 artist styles here:https://mega.nz/folder/FoQR1RgA#gDLJyJe9-JK2_FDaosgSzg

  本模型完全开源不会收费  

  All models can do females, it's just that I didn't specially tweak these models for females

  这是一个带有一点 Niji6 风格的动漫模型(但它可能会失败,这取决于你如何看待它),它不是很 niji6,风格很奇怪

  这个模型有疯狂的颜色和对比度(但有时它太黄了xd)

  ##########################################################################

https://civitai.com/models/89804/ip-design-or-3d

ipDESIGN3D_v31

  为皮克斯风格训练的大模型可以输出皮克斯3D的感觉,类似于泡泡玛特的Q版风格,也可以3D渲染图的2D画面,主要产生偏向Q版的可爱风

  keyword "Meng Meng"

  Generation Suggestion: VAE: cute vae.safetensors Steps:20~30 Recommended: 512*768 (or512*512) Enable the HD Sampler: DDIM

  Words: best quality, amazing quality, great quality, normal quality, bad quality, worst quality,

  aesthetic tags: very aesthetic, Aesthetic, displeasing, very displeasing

  ##########################################################################

  LCM_Dreamshaper_v7

  ##########################################################################

https://civitai.com/models/43331/majicmix-realistic

majicmix7

  不用多说

  推荐参数 V7推荐参数:

  Sampler: Euler a, Euler, restart

  Steps: 20~40

  Hires upscaler: ESRGAN 4x or 4x-UltraSharp or 8x_NMKD-Superscale_150000_G

  Hires upscale: 2+

  Hires steps: 15+

  Hires denoising strength: 0.05~0.5

  clip skip 2

  如果要修复脸部,请使用after detailer.

  如果你的脸画得不好,请在细节处理后使用,而不是面部修复。

  我习惯开启Dynamic Thresholding来更好控制cfg值,1~20都可以尝试一下。

  如果要添加滤镜效果、噪点,请使用BMAB.

  ##########################################################################

https://civitai.com/models/41865/majicmix-fantasy

majicmixFantasy_v30Vae

  不用多说

  Enjoy the randomness!(享受随机性)

  TRY THIS: official art, unity 8k wallpaper, ultra detailed, beautiful and aesthetic, masterpiece, best quality, (zentangle, mandala, tangle, entangle), (fractal art:1.3) , 1girl, extremely detailed, dynamic angle, cowboyshot, the most beautiful form of chaos, elegant, a brutalist designed, vivid colours, romanticism, by james jean, roby dwi antono, ross tran, francis bacon, michal mraz, adrian ghenie, petra cortright, gerhard richter, takato yamamoto, ashley wood, atmospheric, ecstasy of musical notes, streaming musical notes visible

  这个模型风格炸裂,远距离脸部需要inpaint以达成最好效果

  推荐的正面提示recommended positive prompts: official art, unity 8k wallpaper, ultra detailed, beautiful and aesthetic, beautiful, masterpiece, best quality, (zentangle, mandala, tangle, entangle)

  use ng_deepnegative_v1_75t and badhandv4 in negative

  inpaint the face-->only masked-->set to 512x512-->Denoising strength:0.2~0.5

  ##########################################################################

miaoshouai-anything-v4-pruned

未知

  ##########################################################################

https://civitai.com/models/85504/oukaniji5

oukaNiji5_v10

  伪niji5

  V1版主要表现优势为人物,擅长平涂,厚涂,Q版风格,同时保持niji5在颜色上丰富的表现

  本模型推荐使用EasyNegative,会让质量变得更高,

  优化器使用DPM++ SDE Karras,UniPC,DPM++ 2S a Karras

  放大算法推荐使用 R-ESRGAN 4x+

  VAE推荐:animevae,anything-v4.0

  其他设置大家可以自行尝试,欢迎你分享你的作品,这会让我持续迭代改进。

  ##########################################################################

https://civitai.com/models/241415/picxreal

picxReal_10

  真实场景,人物

  The flexibility of the model means that a basic understanding of how to write a prompt is required - describe desired main style, main scene, details, effects and artistic techniques, and in response it will show something pleasant. Everything else is traditionally on the preview.

  ##########################################################################

https://civitai.com/models/208098/rainbowsweets

rainbowsweets_v20

  这个模型可以产生各种各样的内容,虽然它可以产生 NSFW 图像,但我主要关注的不是这一点。

  VAE: I created my own and baked it in.

  Sampling Method: DPM++ 2M Karras, Restart

  Steps: 25 – 30

  Image Sizes: 512x512,512x640,512x768

  CFG Scale: 6-8

  Clip Skip: 1

  HiRes Upscaler: 1x-ITF-SkinDiffDetail-Lite-v1 or R-ESRGAN 4x+ Anime6B

  Hires Steps: 15-20

  Denoising Strength: 0.6

  Prompting:

  Can handle both simple and more complex. Masterpiece and High Quality aren't typically needed from what I've seen but use whatever you like. Something as simple as 1girl can show great results with details and variety.

  Negative:

(Highly recommend Async’s Negatives, Async's Negative - v11 | Stable Diffusion Embedding | Civitai)

  These are just examples of what I’ve used, but I’ve tried many Negative prompts that have good results:

  an11, an10, worst quality, low quality, multiple limbs, multiple legs, bad anatomy, NegfeetV2, bad-hands-5, badhandv4

an10, (sketch, watermark, signature, worst quality, low quality, large head, extra digits:1.4)

an11_light

  ##########################################################################

https://civitai.com/models/7371/rev-animated

revAnimated_v11

  动画:该模型能够创建类似 2.5D 的图像生成。此模型是检查点合并,这意味着它是其他模型的产物,用于创建从原始模型派生的产品

  LoRA友好

  它在以下分辨率维度上效果最佳:约512×512 约512×768 约768x512

  VAE系列:

  orangemix.vae.pt   kl-f8-anime2.ckpt   Blessed2.vae.pt

  ##########################################################################

https://civitai.com/models/65203/disney-pixar-cartoon-type-a

https://civitai.com/models/75650/disney-pixar-cartoon-type-b

disneyPixarCartoon_v10 (1).safetensors

disneyPixarCartoon_v10.safetensors

  There's no good pixar disney looking cartoon model yet so i decided to make one. This is a merge of Pixar Style Model with my own Loras to create a generic 3d looking western cartoon. In the examples 

  I Use hires. fix, upscale latent, denoising 0.5, hires steps 20, upscale by 2 . Clip skip 2 . I will upload more variants of this in the future

  ##########################################################################



Copyright © 2024 aigcdaily.cn  北京智识时代科技有限公司  版权所有  京ICP备2023006237号-1