So... I was trying to generate Pearl with her mother and couldn’t get it to work properly. I spent a lot of time creating references (almost a hundred, which I may still update)
The goal was simple: two girls with the most distinct features possible (size, haircut, skin color, hair type...) interacting together.
The interactions are basic, SFW-but-“sensitive”: kissing, piling up, hugging, kneeling next to each other. The idea was to build a solid reference base that could be useful in both SFW and NSFW contexts.
My objective: create an assistant LoRA able to recognize when it’s time to generate two characters with slightly different appearances, so multi-character generation stops messing up my two OCs.
Did I manage to generate exactly what I aimed for? No. It didn’t really help with generating two existing characters with very distinct designs.
Is this LoRA useless, then? I don’t think so. It actually works well for what it was trained to do: generating two non-existing characters together, handling interactions from simple to more complex ones. And who knows — maybe you’ll get better results than I did when using it to assist with existing characters.
But when it comes to generating two random girls with highly different features, this thing turns out to be pretty useful, so I decided to share it.
You can try prompts like:
2girls, different features, different skin color, distinct appearances, just change whatever you want... haircuts, body types, proportions, etc.
