QQ|Archiver|手机版|小黑屋|AiOMG ( 粤ICP备2022147020号-1 )

GMT+8, 2024-11-25 20:34 , Processed in 0.041497 second(s), 26 queries .

Powered by Discuz! X3.5 Licensed

© 2001-2024 Discuz! Team.

[LoRA] Better eyes+face+skin | 更好的眼睛+脸+皮肤

42
0
0
0
simonlive LV9
正文
发布时间:2024-07-14
模型类型
模型类型: LoRA
基础模型: SD 1.5
最新版本: v2.0
触发词:
下载链接: https://pan.baidu.com/s/133-P5mQUSKxE7W0YbJ9JkQ?pwd=t662
提取码: t662
源链接: https://civitai.com/models/51430?modelVersionId=55905
模型简介:
如果你觉得我的模型有帮助,请点亮右上角那个免费的收藏红心,不胜感激。

both:
· Key feature: Enhanced eye detail throughout the face, rather than just a single eye close-up
· 主要特征:在整张面部进行眼睛细节的提升,而非仅仅是单只眼睛的特写
· It usually makes the face more beautiful and adds details to the skin
· 通常会使人物面部更加漂亮,并增加皮肤细节
· Can be mixed with other Loras for better facial results
· 可以与其他LoRA混合使用,获得更好的面部效果
· Suggested weight: Whatever you want, just don't make it too big
· 建议权重:任何你想要的,别太大就好

v2:
更新了训练集,提高了分辨率,细节更加丰富真实,但按照全图放大四倍的时候会出现面部崩坏,建议仅放大2倍或者用于inpainting。

v1:
细节更加丰富的眼睛,可以用于inpainting,也可以直接生成图片以改善眼睛的细节,效果如封面图所示。建议使用Highres或者img2img放大。以使得效果明显。

因为主要在欧美面孔上训练,故而会略微影响面部,偏向于欧美风格,因此可以变换亚洲面孔的模型。

推荐的扩展:
https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111
https://github.com/ototadana/sd-face-editor

~~~~~~~~~~~~~~~~~~~~~~~o(* ̄︶ ̄*)o~~~~~~~~~~~~~~~~~~~~~~~

主图提示词和参数:
Prompt: (masterpiece:1.2), (best quality:1.2), hyper realistic,highestres,absurdres,extremely detailed,volumetic,dramatic lighting,
1girl,detailed eyes,face,looking to the side,<lora:detailed_eye-10:1>

Negative prompt: verybadimagenegative_v1.3,low quality, worst quality, (bad anatomy), (inaccurate limb:1.2),bad composition, inaccurate eyes, extra digit,fewer digits,(extra arms:1.2),  nsfw,multiple view,border,white background,simple background,plain background,manga,oversaturated, multiple tails,duplicate,artist name, cloned face, grayscale, unclear eyes, wrong anatomy,child,canvas frame,fake,amateur drawing,toddlers,  Unspeakable-Horrors-Composition-4v bad-hands-5

Steps: 20, ENSD: 31774, Size: 2048x2048, Seed: 1274936605, Model: majicmixRealistic_v4, Sampler: DPM++ 2M Karras, CFG scale: 7, 'Overlap': 48, Clip skip: 2, Mask blur: 4, "{'Method': 'MultiDiffusion', 'Upscaler': '4x-UltraSharp', Model hash: d819c8be6b, 'Scale factor': 2, 'Keep input size': True}", 'Tile batch size': 8, Denoising strength: 0.35, 'Latent tile width': 96, 'Latent tile height': 96, Tiled Diffusion upscaler: 4x-UltraSharp, Tiled Diffusion scale factor: 2

本帖子中包含更多资源

您需要 登录 才可以下载或查看,没有帐号?立即注册 微信登录 手机动态码快速登录

x
回复

使用道具

 
您需要登录后才可以回帖 登录 | 立即注册 微信登录 手机动态码快速登录

本版积分规则

快速回复 返回顶部 返回列表