This is my first attempt at training a LORA, I used the Harry Potter Legacy character Mirabel Garlick as I think it was recent enough to not be already included in mainstream Checkpoints, so I hope this can be helpful for your works!

Feel free to share you attempts and feedback!

it needs <lora:Mirabel_Garlick:1> of weight to get the exact details (the hat, the bows and such) I wouldn't go lower than 0.8 or it just lose all it's resemblances.

the keyword is Mirabel_Garlick and I used the Danbooru tag system to train it so it works best using individual tags as prompts but let me know if it works on your descriptive workflow as well.

More pics and outtakes

I have a 1070 so I had to cut some corners while training this model but I hope you can enjoy it! (let me know if you wanna try and merge it with something else because I'm curious of the results :) )