Was trying to train a LORA but without luck, so I collected some images from boorus for experiments, and the outcome looks good to me (pov as a newbie).

Trained on NAI, tested on AOM3.

A weight of 0.6 - 0.9 and some prompts like hair bow should generate OK results.

Somehow it might be a bit over trained and might require additional positive/negative prompts and clip skip change to overcome.

Example prompts: 1girl, bangs, black_headwear, blonde_hair, blush, bow, broom, broom_riding, candy, cape, dress, eyebrows_visible_through_hair, food, gloves, green_eyes, hair_between_eyes, halloween, hat, holding, holding_staff, holding_wand, lollipop, looking_at_viewer, simple_background, stuffed_animal, white_background, white_gloves, witch_hat