Basic Info
I would like a LoRA that can generate both Jinshi and Maomao from the Apothecary Diaries with at least 70% accuracy.
This lora must:
Activate with the trigger words jinmao and/or Jinshi and Maomao.
Be compatible with a variety of anime and 2.5D models.
Output at optimal resolutions 512 x 768 and 768 x 512.
Reproduce a variety of outfits and scenarios.
I don't need the LoRA to accurately reproduce the character's height difference, but it would be a bonus if it could.
Previous Trials
I've tried making this LoRA myself; however, as it was my first, the overfitting was atrocious. I have included my test dataset in the attachments if you wish to peruse.
All sample images were generated using a combo of both my test Lora and alea31415's Apothecary Diaries Character LoRA at very low weights with fiddly prompting. I can get decent output this way, but as stated, it's incredibly fiddly, as the LoRAs are both overfitted.
Higher weights granted more accuracy and reliability when reproducing the character likenesses but tended to reuse the training data's poses and clothing.
Lower weights were more flexible, but the chance of getting one character with the traits of both (or some other form of character likeness bleeding into the other) was extremely high.
I am too inexperienced to know how to fix it, nor am I interested in taking the time to learn. Thus, this bounty.