Hello all. I've been running comfyui a couple of months now and have three favorite checkpoints I use, but want to enhance them. I have some questions about my project. I think it's either going to be a multi character LoRA or a style LoRA, or perhaps a checkpoint merge. Im very unclear as to the best way to approach this.

I like the styles of 3 different checkpoints for different anime type styles but left on their own they seem to have been trained with only one or two faces. I can get diverse skin tones with the correct prompts but not so much with faces. So, I'm thinking of using the faces, in various poses, somewhere around 120 different individuals of various ethnicities, rendered in the style of the model. I'd likely do whatever this project turns out to be 3x, (once for each checkpoint) due to their differing styles.

I've never trained a LoRA before nor created a checkpoint or checkpoint merge, but I think I can trigger a specific individual with use of the right tag, and probably attributes like skin color or ethnicity the same way. Not sure if it will randomly pick an individual or use a mix of features I'm training without a trigger prompt. The checkpoints I use seem to randomize these attributes automatically at the moment but there's not much diversity in features etc.

I'm planning on running the training locally. I want it to be on point with my concept and trained as efficiently as possible because my KWH charge is very high. A couple weeks of 8 hrs/day last month cost me about $60 in higher electric costs, for example.

Thank you in advance. I'm new to this and appreciate your helpfulness and understanding of my novice status and the distinct possibility I'm looking at addressing my concept incorrectly.