写两次1girl还是2girls呢？人名直接放在2girls后面写两个经常会人物特征融合啊

动作那部分用2girls，角色那部分用1girl，这个我用模型自带的tag效果还可以但用lora就得看lora质量了，过拟合模型你怎么写tag都不行，另外咋哪里都能看到你啊

光辉没办法，我是先写2girls，然后接其中一个girl的特征，然后写动作，背景等等，最后写另一个girl的特征，总之就是尽量把两人的特征隔远点，这样污染的几率稍微低点，当然最终还是得抽卡。。。
比如这个：（抽卡难度还是太高，万分建议分区提示词或者局部重绘）
anime coloring, 2girls, ellen \(touhou\), touhou, 1girl, yellow eyes, hair between eyes, blonde hair, long hair, red hairband, white shirt, red skirt, red vest, waist apron, white apron, puffy short sleeves, red bow, buttons, cowboy shot, side-by-side, standing, arm hug, v, dynamic pose, smile, detailed background, outdoors, soft natural lighting, sky, cityscape, inu sakuya \(nejikirio\), touhou, 1girl, maid, maid headdress, dog ears, dog tail, grey tail, blue eyes, grey hair, short hair, twin braids, ahoge, black shirt, black skirt, long sleeves, white vest, masterpiece, best quality, <lora:!SDXL\sdxl_hyper_sd_8steps_lora:0.80>
Anima就是直接自然语言写了，只要是在不同自然语句并且句号隔开了，污染就很小了。
比如这样：这个抽卡就容易很多了。。。
anime coloring, Two anime girl standing side-by-side outdoors, with dynamic pose. The girl on left is inu sakuya \(nejikirio\), touhou, 1girl, maid, maid headdress, dog ears, dog tail, grey tail, blue eyes, v, grey hair, short hair, twin braids, with ahoge. Another girl is ellen \(touhou\), touhou, 1girl, yellow eyes, hair between eyes, blonde hair, long hair, red hairband, arm hug, white shirt, red skirt, red vest, waist apron, white apron, puffy short sleeves, red bow, buttons. The background is outdoors, cityscape, sky, tree, building, with soft natural lighting. detailed background, detailed face and eyes, masterpiece, <lora:anima\anima-turbo-lora-v0.2:0.80>
没有对比就没有伤害，Anima的加速lora损失的画质还是太多了。。。

Anima这边白毛少写了几个特征，导致衣服不太一样，重新跑一下。特地挑的这两个角色引用都只有100多条，但是好像两边的形象还不一样，两个都不认识，我也不知道哪边更还原。。。然后Anima大的特征倒还好，小的特征还是很容易串，比如白毛的呆毛就很容易串到金毛头上。。。
anime coloring, masterpiece, best quality, score_8, highres, absurdres, Two anime girl standing side-by-side outdoors, with dynamic pose. The girl on left is inu sakuya \(nejikirio\), touhou, 1girl, maid, maid headdress, white vest, black shirt, black skirt, long sleeves, dog ears, dog tail, grey tail, blue eyes, v, grey hair, short hair, twin braids, with ahoge. Another girl is ellen \(touhou\), touhou, 1girl, yellow eyes, hair between eyes, blonde hair, long hair, red hairband, arm hug, white shirt, red skirt, red vest, waist apron, white apron, puffy short sleeves, red bow, buttons. The background is outdoors, cityscape, sky, tree, building, with soft natural lighting. detailed background, detailed face and eyes, <lora:anima\anima-turbo-lora-v0.2:0.80>

Regional prompt最好用上，不能用角色Lora

光辉noob的话文生图双胞胎然后图生图重绘其中一个（需要openpose控制）
Anime 可以直接出
2girls,
souryuu asuka langley, neon genesis evangelion, orange hair, twintails, red hairband, blue eyes, tokyo-3 middle school uniform, white shirt, red bow, black skirt, kneehigh socks, confident, tsundere, grin,
ayanami rei, neon genesis evangelion, blue hair, bob cut, short hair, red eyes, tokyo-3 middle school uniform, white shirt, black skirt, kneehigh socks, expressionless, looking away,
back to back, standing, outdoors, sunny day, school courtyard, cherry blossoms, (medium shot:2)
The composition is a medium shot of Asuka and Rei standing back-to-back in a sunlit school courtyard. Asuka on the left is grinning confidently with her arms crossed, while Rei on the right faces away with a calm, expressionless look. Warm afternoon light filters through cherry blossom trees, casting soft shadows on their school uniforms, with scattered pink petals drifting in the air and a soft-focus school building in the background.

2girls,
kirisame marisa, alice margatroid, touhou,
blonde hair, twintails, yellow eyes, witch hat, hat bow, red bowtie, black dress, white apron, waist apron,
blonde hair, long hair, blue eyes, blue dress, white apron, red hairband, frilled apron, frilled dress,
heart hands duo, heart hands, smiling, looking at viewer, standing,
(upper body:3)
The composition is a medium shot of two girls standing side by side, forming a heart shape with their hands together in front of them. Soft, warm lighting illuminates their cheerful faces, with a bright and airy atmosphere and a gentle pastel color palette that enhances the cute and affectionate mood.

那么按照7楼的写法……3楼的prompt应该是
2girls,
inu sakuya \(nejikirio\), ellen \(touhou\),
maid, maid headdress, dog ears, dog tail, grey tail, blue eyes, grey hair, short hair, twin braids, ahoge
yellow eyes, hair between eyes, blonde hair, long hair, red hairband, white shirt, red skirt, red vest, waist apron, white apron, puffy short sleeves, red bow, buttons
locked arms, v, double v, looking at viewer, smile, blush, standing, outdoors, sunny day, blue sky, clouds,
The composition is a medium shot of two girls standing close together, arm in arm, both facing the viewer with bright smiles and cheerful expressions. Soft sunlight falls from above, casting gentle shadows, with a clear blue sky and fluffy white clouds creating a lighthearted, whimsical atmosphere.

很明显的一主一副就1girl。很明显的两主角就2girls。模糊的我的做法是不输入girl。

@503ycy
文生图，模型是某光辉
kirisame marisa,touhou,2girls,standing,heart hands duo,cuddling,outdoors,flower,blue sky,
多生成几张选合适，选的是方便替换的+controlnet预处理器方便识别的（所以拥抱这类交互性更强就很麻烦）
选了左下，左边没帽子的那个适合被替换~~右下也不错人物更大方便预处理器识别。如果左下识别困难就用右边
图生图局部重绘。在这里提示词就要改成你需要换的人物了，这次选了帕秋莉
patchouli knowledge,touhou,2girls,standing,heart hands duo,cuddling,outdoors,flower,blue sky,
然后涂需要重绘的部分，建议涂大一点，因为不同角色发型服装身材什么的不一样。背后的手有点讨厌也涂掉
红圈是改动过的参数，没圈的就是不动的
先不用openpose跑一次。其实光辉很强了，是有不错的局部重绘能力
帽子可能不对，看来重绘范围还可以再扩充一点
OPENPOSE相关
左边人物背后那只手识别出来了。编辑去掉
生成~~
生成出来的图仔细看可以看到我涂抹重绘的“接缝”。这种“接缝”在图生图（非重绘）中低重绘幅度（我一般是0.4）再跑一次可以消除。提示词用WD反推的就行了