Ever wanted to put loads of characters in in image, but all of your people blend into some hybrid monster thing or all look exactly the same?
Or did you want septuplets in your images?
Here's a brief overview with examples can hopefully help you get your desired results more frequently.
SD-WEBUI FORGE:
Required Extensions:
Attention Couple - this is basically AUTO1111's regional prompting, but easier to use
Optional Extensions:
Reactor - recommended but not required; used for face consistency (kind of a pain to get working for some - me being that "some")
ControlNet Models - recommended but not required; if you give ControlNet a mask with 2 people in it, you're far more likely to get 2 people in the resulting image(s). I use depth, but it's your choice - a lot of folks seem to like IPAdapter
sd-forge-couple is the only tool you really need. Once installed, your TXT2IMG & IMG2IMG tabs include the Forge Couple collapsable menu and look something like this:
The Forge Couple Github page has some great documentation on what all the radio buttons do and whatnot, so I'll just explain how I use it.
In the interest of time, I'll begin with the picture below - and toss into the IMG2IMG tab. In case you can't see the images prompt, here's what I used:
[realistic color manga hentai art, inspired by (Sekirei), forbidden love theme, realistic characters and environment, two distinct characters with unique physical characteristics, realistic natural color theme, high contrast, two beautiful women holding each other in a loving embrace, looking into each other's eyes, Musubi and Kazehana from Sekirei, panorama, close-up, lips focus], [in the style of Ninja Scroll and Naughty America, fighting each other in the Sekirei Plan, MBI headquarters in Teito Tower immediately after an explosion caused the building to collapse, thick dark smoke now rises from the tower's foundation in the background, outdoors during the day, metropolitan cityscape, urban environment, the two women fight for their lives in the busy city streets, dangerous, destructive, both women are super-powered humanoids with unique powers who must fight in a battle royal called the Sekirei Plan, staring at one another, dynamic fighting action poses, direct sunlight illumination, volumetric smoke and debris, elongated shadows, poster for an action and adventure movie, glossy finish] <lora:H001PXL:0.4> #
[Character1- pictured left: (Misubi), a beautiful young adult woman with short brown hair, greatly enhanced strength, reinforced by heavy combat gloves, small breasts, braless, covered nipples, short red skirt, tattered and torn short white robes of a shinto priestess, white cloth belt loosely tied in front, cleavage, innocent, submissive, defensive posture, taunt and pump kick, sexy, clothing in tatters from fighting that reveal her sexy curves, Cathected,, Rejecting, "Airheaded Glutton", the spirit of a ferocious black bear appears behind her and fights with her] <lora:SlimesMusubi:1> #
[Character2- pictured right: (Kazehana), tall, extremely long black hair, buxom adult woman, very large breasts, cleavage, extremely tight-fitting long purple split-leg dress with a hemline that ends just below the tops of her thighs, her top is laced together in front and the laces are struggling to contain her massive bust, bare long sexy legs, bare empty hands, "Obscene Exhibitionist", braless, dominant, offensive attacking posture, sidestepping and counterattacking, sexy, miserliness, Frustrated, spends most of her time lounging and drinking sake but today she is filled with bloodlust and rage, member of the First Disciplinary Squad, ability to control and manipulate wind, can knock down opponents and fly short distances using the wind] <lora:SlimesKazehana:1> <lora:SlimesKazehana:1> #
HIGHLIGHTED TEXT: I use wildcard values for these words, really looks like prog420ress/fighting, {2$$prog420ress/emotion_tense__ in the prompt
This is a prompt for Flux, so there's no negative prompt input. I like to format prompts like this to keep them tidy, just make sure you don't use \n (newline/enter button) until you want to change characters. — Char1 blahblah <lora1> \n Char2 moreblah <lora2>
If you're using the Advanced option, be sure to have one region/layer that ecompasses the entire image This is your "global" layer, and will be applied to everything So if you're using Pony, that'd be where you put your score_7_up and whatnot. Since I'm using Flux, I kind of just write a little story to set up the environment. Be sure to mention the number of characters in your image and if your characters have names, go ahead and use them!
ie: two beautiful women holding each other in a loving embrace, looking into each other's eyes, Musubi and Kazehana from Sekirei,
(Hmmm…. Something tells me that's not the prompt I used for the image, but that's okay - I'll use it for this example!)
NOTE: If you change the prompt at the top of the page, it will alter the prompts shown in the screenshot. So before you begin, make sure your region boxes and character prompts match up!
So now I have some example images and, more importantly, my prompt is embedded in their metadata, so I'll load one that I'm happy with into the PNG Info tab and send it over to TXT2IMG.
(the prompt reads 'loving embrace'... yeah, maybe they're looking a bit more friendly toward each other now)
In TXT2IMG, I'm going to swap out Slime's Kazehana with my Arbiter, define new regions, add some wildcards and we could use ControlNet and Reactor for consistency as well as fresh fighting positions, but I'm not going to this time around (if you'd like to know how I use these tools, let me know and I'll see about making more of this kind of article)
Toss the results into IMG2IMG for any positions or poses you like, and you've got a method in which you can change up your characters and positions all in WEBUI-Forge:
Arbiter steals Slime's Musubi's gloves in all of the above pictures. If I add "bare hands" to the Arbiter's regional prompt, she'll stop taking Musubi's gloves since I will have told her she isn't wearing gloves.
This goes for every aspect of the characters - Musubi isn't stealing the Arbiter's helmet because I defined her hairstyle in the regional prompt.
Hair color, eye color, tattoos, footwear, anything, really… if one character has them and you don't mention the other does not - they'll share them
Where Musubi's character got that ponytail in IMG#1? Your guess is as good as mine, but it happens..
SD-WEBUI-Forge's Attention Coupler works 100% of the time about 50% of the time Ha!
If you include a few more tools, you'll get more unique characters that are accurately represented in your images more frequently - but it's not guaranteed.
The most reliable way to ensure the only character affected is the one you intend to is through inpainting, but I bet you already knew that!