Why is firefly so bad? local image generation is better yet im paying $70 a month for this

I am trying to get used to firefly after not having used it for about a year and a half and it is honestly astounding how bad it is considering how far generative AI has come elsewhere. Firefly has infact gone backwards it seems, as now, constantly and arbitrarily it refuses to generate something for breaking its completely nonsensical guidelines, for example, trying to generate something of a fantasy image - it will constantly block generation of anything deemed 'menacing' or 'threatening', then randomly allow it, if someone is paying this much a for image creation software, you would assume it would be less restrictive than the countless free alternatives out there that can even run without connection to the internet. Secondly I could've sworn it was far more intelligent when it comes to using generative fill to normalise an element of larger compositon, i.e created a background and foreground then selecting forground elements and either typing nothing into the generative fill prompt or writing something along the lines of "match background lighting and tone only, do not add or remove objects" will always either completely remove the selected element or replace it with some entirely random other object. mid last year when i first starting using firefly, i was generating whole landscapes with generate image then generative fill on elements until it was exactly what i wanted, now generate image usally doesnt work, so ill use stable diffusion to get the base image, then using generative fill it will just randomly mutilate selected parts of the image until i give up and go to some other generative AI tool only to bring it back into photoshop for resizing and small corrections. I feel like i must be using it incorrectly because i can't believe adobe would flaunt and market itself on the basis of a tool that is so downright awful and far behind industry standards. I have tried to find a defined syntax for how to word prompts but only find generalisations and suggested techniques that do not work at all on adobes tutorials  (telling me to select space between fore and background to smooth the transition between them - again creates a mess of mutilated images woven together) It is just far too convenient having AI generation in the exact same place as the rest of the photoshop tools and i am really sick of going Dall.e, stable diffusion, midjourney > photoshop, then back to extrernal AI tool, back and forth ad nauseam.