Model is trained on captions (full sentences) and doesn't work as well with tags (like flux in general). Best results come when referencing the full name and giving a bit of description like:
Faith Connors the main character of iconic game Mirror's Edge is doing ...