How To Make Multiple Characters
This guide will teach you about all details, useful tools and tricks to create perfect couple images including suggested models, LoRAs and effective prompting.
Last updated
This guide will teach you about all details, useful tools and tricks to create perfect couple images including suggested models, LoRAs and effective prompting.
Last updated
● Use couple tag before your subjects and add hetero-yaoi-yuri depending on what type couple you want to generate.
● Using duo focus is effective at avoiding pov views and generating both subjects at same image.
● More characteristic term in prompt cause more randomized results, try to choose model that recognize your character or use LoRA for it and keep characteristics simple. Avoid to use many terms like black hair, green eyes, blue hair this type terms can effect both of subjects or wrong one without BREAK, but remind that BREAK is not 100% working technique.
● Use 2boys instead of 1boy, if you are going to make yaoi(gay) couple. | Use 2girls instead of 1girl, if you are going to make yuri(lesbian) couple.
● Use duplicate, bilateral symmetry in negative prompt to avoid similar characters.
● Let's say i want to create Mario x Ahri couple, i must start write like this for my subject: couple, hetero, duo focus, 1girl, 1boy
● If i want to create Makima x Sakura couple: couple, yuri, duo focus, 2girls
Note: Prompting can show difference depending on model you use, this is only an example based on Danbooru tags and Model's general training.
● Most of anime models has pretty small character memory because of their training.
● Generally models recognize characters from most popular Danbooru Character Tags and also some of quite popular characters such as Mario, certain league of legends characters, protagonist of popular games etc.
● For detailed and accurate couples, LoRAs become important part of generation.
● BREAK keyword is effective for keeping character's characteristics seperate and don't blend them.
Positive Prompt
(masterpiece, best quality:1.2), highly detailed, (couple), hetero, (duo focus), 1girl, 1boy, holding hands, looking at another, beach scenery, bokeh, (depth of field:1.2), absurdres, highres, soft lighting, soft shadows, BREAK, 1girl, nami (one piece), NamiFinal, ginger, long hair, pale skin, busty, bikini, BREAK, 1boy, uzumaki naruto, blonde, short hair, shorts
Negative Prompt
EasyNegativeV2, (badhandv4), nsfw, duplicate, (blending), (bilateral symmetry:1.5)
LoRA
Nami (ナミ) One Piece Character LoRA (Post-timeskip)
Naruto Uzumaki / 漩涡鸣人 / ずまき ナルト / 火影忍者
Embedding
[TI] EasyNegativeV2 [Textual Inversion Embedding]
badhandv4
Model AnythingElse
Explanation
As you can see i used BREAK here and made 3 seperate prompt to avoid blending.
Part: General prompt, you must use both subjects here otherwise models will create single subject. I used (couple), hetero, (duo focus), 1girl, 1boy for subjects and add my additional prompts.
Part: Name of first subject and details, i used 1girl for first subject and add her characteristics and LoRA to make her look like Nami from One Piece.
Part: Name of second subject and details, i used 1boy for second subject and add his characteristics and LoRA to make him look like Uzumaki Naruto from Naruto.
You can create ship images with LoRAs and BREAK but i have to admit it, it's hard and not worthy.
Most of SD1.5 Models towards to generate single subject and you can't even get 2 person in same image.
Prompt understanding is not that strong, even with BREAK there is a still chance to get mixed characteristics. BREAK is not 100% working trick.
There is high change to get bad and random compositions.
They have bigger character database compared to SD1.5 Models and creates more accurate characters.
Generally models recognize most of characters from Danbooru Characters Tags and many well-known characters from different genres such as movie, anime, game etc.
LoRAs are not necessary but might be useful at some creations.
Animagine XL is very great choice to make ship images, it's better than other SDXL based models and has larger character database.
Positive Prompt
couple, duo focus, hetero, 1girl, 1boy, boa hancock hugging to Luffy, one piece, ship background, lovely, masterpiece, best quality, safe, SFW, very aesthetic, recent, absurdres, highres
Negative Prompt
(worst quality, low quality:1.3), NSFW, explicit, oldest, early, (very displeasing, displeasing), lowres, bad anatomy, anatomical nonsense, bad hands, jpeg artifacts, ((ugly)), grayscale, traditional media, duplicate
Model Animagine XL
It's overall good, especially Animagine XL model is so good at recognizing characters, this means you don't have to prompt details and there will be no blended characteristics most of time.
Pony models are actually SDXL based and this means they are atleast good option as SDXL models.
They are very customizeable so you can create almost every type couple you can imagine.
Generally recognizes most of well-known characters from different genres including anime, game, cartoon, comics etc.
AutismMix is good for anime couples while T-Ponynai3 especially excels at Genshin Impact characters.
Positive Prompt
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, (anime), sfw, couple, yuri, duo focus, 2girls, hu tao (genshin impact), eula (genshin impact), carrying another on arms, hugging, lovely couple, long wedding dress, wedding suit, wedding ceremony, outdoors, (nature)
Negative Prompt
(score_4, score_3, score_2, score_1), ugly, bad anatomy, bad hands, nsfw, duplicate, blending
Model T-Ponynai3 V6
It's overall good and has various options for different concepts.
Characteristics can blend sometimes depending on what you try to generate.
Creative and impressive results.
Good at rendering details
