>>24868
>there some trick to having two people on a picture
At the current moment in time (and in the past, really), it is a tad tricky. So the feature first came onto the scene as 'Latent Couple', an extension that allowed what was essentially two generation "spaces" within one generation. Later came more extensions, namely Composable LoRa, which allowed using separate loras in each generation space. The thing is, the implementation is arcane and buggy and stupid. So even more extensions came along, like Regional Prompter, that tried to simply the process, but ALL were shitty and buggy and had limitations.
That brings us to today. Now, the only extension I can even halfway recommend is Forge Couple. This requires you to be using SD Forge, so that's already annoying. Doubly annoying is that it's not listed on the available extensions within the UI: you have to install from URL: https://github.com/Haoming02/sd-forge-couple
However, this extension reduces all the fluff and stupid pomp of the original extension and boils it all down: each new line in your text prompt is a region. You can set the first or last line can be defined as the "global effect" of the image. Basically composition, style, background, etc. You can see that in the meta data for many of the un-edited two-girl images I've posted. You then repeat a small part of the overall prompt in each region prompt, so if you have "2girls, yuri, bedroom" in the "global effect", you would still put "2girls, yuri, blonde" and "2girls, yuri, brunette" for your other two regions.
After you set everything up, be prepared for a lot of trial and error. Did you use too many loras? Black screens. Did you not use the right weights for those loras? Over-saturated mess. Did you use high-res fix or adetailer? Be prepared for errors about your sampler choice, or other arcane mumblings. ACTUALLY get it to generate correctly? Savor that feeling, because you can literally generate the same thing again and it won't work, with no changes, until you close and re-open the console. It's the most temperamental stupid piece of crap. But, when it works, it works pretty ok.
Easiest: Each character is already defined in the checkpoint model and no character loras are required, only styles.
Medium difficulty: Both characters are defined in the same multi-character Lora OR one character is defined in the checkpoint and one requires a lora.
Insane difficulty: Both characters require separate loras. This will almost never work, ESPECIALLY with hi-res fix or adetailer.
Inevitably, especially if your prompt implies the subjects are close together, things will be merged a bit sometimes. For example, you can see in the attached Melony's sleeve/top took on the black appearance of Cynthia's wardrobe, because they were that close in frame. You can try adjusting the Global Weight setting, but this will just happen sometimes.