My 4th try at it and i wont probably try this anymore, this was done on Hollowstrawberry's lora trainer: https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Lora_Trainer.ipynb

Trained with 61 images at 768 resolution for 5 repeats and 10 epochs, used flip_aug and min_snr_ganma, 2e-4 unet/1e-4 text econder lr, batch size 2.

Getting waai fu right is a nightmare and i think part of the problem is the models not being good at it, i only tested with 7thanimev3c: https://huggingface.co/syaimu/7th_Layer/blob/main/7th_anime_v3/7th_anime_v3_C.safetensors but if someone has a merge with a furry model and it looks better on those please do tell me.

More ammount of time went into pruning this time around, i decided to swap 1girl for furry female since leaving it before was detrimental, other than that the minimum tags to prompt her are "waai fu, furry female, tiger ears, body fur", if you want her glasses add "orange-tinted eyewear, round eyewear" and for her default outfir use "default china dress, capelet, pelvic curtain, elbow gloves, orange belts", add or remove "animal nose" to positive/negative prompts depending on what you want. Additionally you can use tiger girl, tiger tail to help prompt the tail but expect random tails around and tail limbs, i honestly dunno how to fix those.

Other than that here is the tag frequency for anyone interested: https://files.catbox.moe/0a76lr.txt