Tags:

This LORA doesn't need any nodes to generate characters it was trained on together.

To combine characters from different LORAs "Regional Prompting" is used, although I haven't used this technique personally yet.

There is good guide for Forge - https://civitai.com/models/339604/how-to-generate-multiple-different-characters-mix-characters-andor-minimize-color-contamination-or-regional-prompt-adetailer-and-inpaint-or-my-workflow

And someone have posted some workflow for ComfyUI, but i don't know if it is any good - https://civitai.com/articles/9534/regional-prompting-in-comfyui.

By the way NoobAI https://civitai.com/models/833294?modelVersionId=1116447 knows thousands of different characters and can generate any of them together without needing special tools.

@lwander You don't necessarily need same amount of images, but amount of steps you train them for should be about the same, it's achieved using repeats (if you have 10 images of first character and 5 images of second character you use 2 repeats for second one). I also have separate folder for images where these characters are together. You can look at metadata of this LORA using this tool https://civitai.com/models/249721, it shows a lot of information including folder structure used for training.

@lwander This is just for outfits to be more flexible, you don't have to tag it. They wear the same outfit, so there is no problem using same tags. If their vest differed, but still had the same color, "green vest" could potentially become some mix of them, although I haven't tested it, maybe model would figure out that when you generate particular character they wear that particular green vest. For multi-character LORA it might be beneficial to not tag their outfit, because in this case character tag will be associated with their outfit, but you will have a hard time generating them with outfit different than their default one in this case, potential solution if your character has different outfits is to balance them and tag like outfit1, outfit2, then it might me able to separate character from outfit, because outfit will be associated with this tag - "outfit1" and not character tag, but if your character wears only one outfit it won't work.

@lwander Also these "|||, brown hair, indoors, green vest, white shirt, formal, milliemv, short hair, sharonmv, black hair" are not concepts, but tags in order of frequency.

@lwander And for these type of LORAs Prodigy might be the best optimizer, I trained this LORA second time for experiment using "CAME" and it mixed characters more often, but i haven't really tested it much. My other multi-character LORA was also trained using Prodigy - https://civitai.com/models/1169259/devola-popola-nier-automata-ver11a