全网首发 | 写实摄影 XSM+Lora Girl-Photographic

先看说明！！！

qq群号:301188613 也请多多关注抖音账号：lora girl。

闲聊：大家好，我是小沙弥，秋名山老司机，开车从来不减速。

（XSM就是我昵称的缩写，lora girl是我愚蠢的小宝贝）

QvQ这是我夏姬八乱搞的大模型，希望大家喜欢。

因为电脑配置原因，跑不出最佳效果，示例图将就看吧,让别人也帮忙跑了几张。包含一些涩涩内容，素材丰富哦！涩图什么的就不方便发了。

特别感谢我的朋友：ℳ๓₯㎕.

特别鸣谢作者：【chocolate 、Showmaker11】 等模型作者

欢迎大家积极返图或者入群以及加抖音交流，谢谢！

例图参数目前大部分填写了.

————————————————————————————————————————————————————————————————————————————————

说明：因为上传太费劲，我上传的模型是FP32的修剪版本。原始版本在这里：liblibAI.

模型内的素材还算比较丰富：人物摄影、香车美女、雄伟建筑、秀丽风景、精密机械、战舰航母、人文美食等等无需其他lora可直出。

1.低配显卡推荐启用ADetailer这个修手修脸插件，我也用了的。

2.关于正面提示词的说明：请尽量用最简单简短的提示词，不需用一大堆的画质词来增加效果，本模型已经调试好画质效果，你甚至可以不用写画质词，比如你只需要简单的写：1girl,full-body,这两个提示词或类似的单词，都能出不错的效果。

3.负面提示词推荐启用嵌入式文件：EasyNegativeV2, BadNegAnatomyV1-neg, badhandv4,Negfeet-neg等。

4.推荐参数：

•vae：vae-ft-mse-840000-ema-pruned 、vqgan_cfw_00011_vae_only.

•采样：DDIM, DMP++2M SDE, DMP++2M SDE Karras, DMP++2M Karras.

•步数：30~50

•高分辨率修复：20步以上

•放大算法：4x-UltraSharpSP+8xPSNR+8x_NMKD-Superscale_150000_G等搭配使用。

•重绘幅度：0.2~0.5

•基础分辨率：512*768、960*512等，横屏反过来。

•放大倍数：1.5倍以上

•cfg：7（有一个CFG插件可以增加细节，Dynamic Thresholding 懂的都懂）

5.推荐lora：

add_detail、Detail Tweaker、GoodHands-beta2、PAseerCloudV1,光泽调节器/Gloss_Tweaker/光沢調整器。

6.推荐的人物lora：小怡同学。(请加Q群获取)

7.推荐embedding嵌入式文件：

EasyNegativeV2, BadNegAnatomyV1-neg, badhandv4, ng_deepnegative_v1_75t,Negfeet-neg。

————————————————————————————————————————————————————————————————————————————————

Attention:

QQ Group Number: 301188613

Also, please follow the TikTok account: lora girl.

Casual Chat: Hello, everyone. I'm Xiaoshami, a skilled driver on the Autumn Mountain.

I never slow down while driving. (XSM is the abbreviation of my nickname, and lora girl is my silly little sweetheart.)

QvQ This is actually a model I casually and randomly put together. Due to my computer's configuration, I couldn't achieve the best performance.

So, please bear with the sample images. Some friends helped to run a few images as well.

contains some ??? content, rich material oh! The ??? picture or something is not easy to send.

PS:Special thanks to my friend: ℳ๓₯㎕.

You are welcome to actively contribute images, join the group, or exchange on TikTok. Thank you!

————————————————————————————————————————————————————————————————————————————————

Notes: Because uploading is too tedious, I uploaded the FP32 pruned version of the model. The original version is here: liblibAI.

The materials included in the model are quite abundant: character photography, luxury cars, beautiful women, magnificent architecture, picturesque scenery, precise mechanics, battleships, aircraft carriers, cultural delicacies, etc. No need for other lora, can directly generate them.

1. For low-end GPUs, it is recommended to enable the ADetailer face and hand tuning plugin that I also used.

2. About frontal prompt notes: try to use the simplest and shortest prompts. No need to use a bunch of quality words to increase effects. This model has been tuned for quality, you can even not use any quality words. For example, you only need simple prompts like: 1girl, full-body. These two words or similar can already generate decent effects.

3. Recommended negative prompts: enable embedded files: EasyNegativeV2, BadNegAnatomyV1-neg, badhandv4, Negfeet-neg, etc.

4.Recommended parameters:

- vae: vae-ft-mse-840000-ema-pruned, vqgan_cfw_00011_vae_only.
- Sampling: DDIM, DMP++2M SDE, DMP++2M SDE Karras, DMP++2M Karras.
- Steps: 30~50
- HR fix: 20+ steps
- Upscaler: 4x-UltraSharpSP+8xPSNR+8x_NMKD-Superscale_150000_G etc.
- Editing extent: 0.2~0.5
- Base resolution: 512*768, 960*512 etc., portrait flipped.
- Upscale multiple: 1.5x or more
- cfg: 7 (There is a CFG plugin to add details, Dynamic Thresholding. Those who know will understand.)

5.Recommended lora:add_detail, Detail Tweaker, GoodHands-beta2, PAseerCloudV1, Gloss_Tweaker/光沢調整器.

6.Recommended character lora: 小怡同学. (Please join the QQ group to get it.)

7.Recommended embedding files:EasyNegativeV2, BadNegAnatomyV1-neg, badhandv4, ng_deepnegative_v1_75t, Negfeet-neg. 

————————————————————————————————————————————————————————————————————————————————