Personally I don't really use any activation tags, there are none in the training. I order my prompts like the animagine announcement advises (1boy/1girl, what character, from which series, everything else in random order*). I am just getting started with SDXL training so there are probably things that could be improved, i'll probably retrain the model some time when i know more.

Here is an example of a prompt i might use:
Positive:
(aqua \(konosuba\), kono subarashii sekai ni shukufuku wo!:1.2), blue eyes, blue hair, hair ornament, long hair, large breasts, 1girl, black necktie, black pants, closed mouth, collared shirt, double v, jacket, looking at viewer, necktie, pants, red background, red jacket, shirt, shirt partially tucked in, shirt tucked in, simple background, sleeves past wrists, smile, solo, v, white shirt, (masterpiece, exceptional, best aesthetic, best quality, masterpiece, extremely detailed:1.1), newest, late

Negative:
(lowres, bad anatomy, bad hands, missing fingers, extra digit, text, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts:1.15), (signature, watermark, username, gumroad username, patreon username, artist name:1.1), multiple views, error, blurry

@Korewaai Honestly no idea, like it says in the description i've only tried it with animagine 3