[AI]用LatentCouple與ComposableLora來做貼貼

ControlNet AI版已經有就不額外介紹了

這次主角是這兩個
https://github.com/opparco/stable-diffusion-webui-two-shot

https://github.com/opparco/stable-diffusion-webui-composable-lora

最近蠻熱門的主題

https://i.imgur.com/efpO5Ji.png

能像這樣對圖片中的區塊分區Diffusion

像這個分三區 那關鍵字就換行之後用AND隔開

例如

2girls
AND silver hair
AND black hair

這樣使用

而ComposableLora則是以AND隔開Lora作用範圍

例如2girls AND <lora:a:0.8>, silver hair AND <lora:b:0.8>, black hair

兩者可以同時作用

簡單介紹完之後就是使用例子了

首先是不套Lora的

一開始用Prompt配合Latent Couple生線稿
rough sketch,outline, (monochrome), 2girls hugging, swoop, long white hair,
(empty white background)
AND rough sketch,outline, (monochrome), 2girls hugging, (flat breasts
loli:1.3), (long white chemise:1.3), swoop, bare shoulder, bare neck, long
white hair, smiling,(empty white background), (rough sketch,outline:1.2),
(monochrome:1.4)
AND rough sketch,outline, (monochrome), 2girls hugging, (big breasts
lady:1.2), (long white chemise:1.3), bare shoulder, bare neck, long white
hair, close mouth, (empty white background), (rough sketch,outline:1.2),
(monochrome:1.4)

https://i.imgur.com/yHiLnda.png

可以看到一次選多個區域的話

divisions 用1:5切成1*5的Array
選取時用0:0-2這樣選取左邊40%的空間

第一步結果
https://i.imgur.com/raGvpXz.png

再來丟進ControlNet
https://i.imgur.com/yVLiNnp.png

得到線稿

第二步上色+一些細節

Prompt:
rough sketch,outline, (monochrome), 2girls hugging, swoop, long hair, (emptywhite background)
AND rough sketch,outline, (monochrome), 2girls hugging, (flat breasts
loli:1.3), (long white chemise:1.3), swoop, bare shoulder, bare neck, long
(white:1.2) hair, smiling,(empty white background), (rough
sketch,outline:1.2), (monochrome:1.4)
AND rough sketch,outline, (monochrome), 2girls hugging, (big breasts
lady:1.2), (long white chemise:1.3), bare shoulder, bare neck, long
(black:1.2) hair, close mouth, (empty white background), (rough
sketch,outline:1.2), (monochrome:1.4)
AND (blue eyes:1.3)
AND (red eyes:1.3)

https://i.imgur.com/gw2ulXK.png

這裡將整張圖片分五步製作

其中有使用
https://github.com/Zuntan03/LatentCoupleHelper
來幫忙選擇區域

也可以使用AI版上那篇分享可能更好
https://www.ptt.cc/AI_Art/E.SPa2r2j5e-F4

成品:
https://i.imgur.com/LB5nJfY.png

https://i.imgur.com/6nIhhmd.png

https://i.imgur.com/LplUhCA.png

https://i.imgur.com/Qscx44L.png

還有一張我很喜歡但是手爛掉的

https://i.imgur.com/RhYbLfx.png

自己做貼貼+撞奶很簡單對吧 :)

再來是使用Lora的簡單例子

一樣先Prompt生構圖
https://i.imgur.com/E0XnYGA.png

https://i.imgur.com/MdvgQlI.png

ControlNet

Prompt:
symmetrical,straight-on, solo, 1girl sitting on the floor, feet apart, empty
white background
AND <lora:re0_emilia:0.8> , 1girl sitting on the
floor, (purple eyes),masterpiece, empty white background
AND <lora:re0_echidna:0.8> , 1 echidna (re zero)
sitting on the floor, black dress, black eyes,masterpiece, empty white
background
AND masterpiece, empty white background

https://i.imgur.com/0fXpcha.png

Latent Couple簡單的設定

https://i.imgur.com/IObxYny.png

結果圖

也能做出這樣把兩個角色貼再一起(物理)的圖

以上就是LatentCouple和ComposableLora的簡單介紹

一起來使用吧 :)

另外用這個要注意一下生一張圖時間會變很久...

--
某天，神父從教堂順路載一位年輕修女回家，神父趁換擋時把手放在年輕修女的膝蓋上，

年輕修女靠近神父的耳邊輕聲說：

『TEmPTaTiON』
--

※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 36.231.97.154 (臺灣) ※ 編輯: clothg34569 (36.231.97.154 臺灣), 03/06/2023 21:48:37