Help me this model issue
I checked out the images you posted there, it kinda looks like you have a character lora or embedding enabled. Could you post the PNG metadata from one of those images? Also, check your embeddings folder for anything with a common word as the filename, like "1girl". I once downloaded an embedding here that the creator named just LOP, and it took me awhile to realize that it was being enabled whenever I had "wlop style" in my prompt.
It's hard to say where you might be going wrong without being able to see the metadata from some of your generations.
Generally, things like
photorealistic, lifelike, stock photo, will tend towards producing realistic images. I've also seen users prompt specific camera models and apertures, but this seems hacky and not a great solution.
Also adding things to the negative prompt such as
anime, drawing, cartoon, render tends to reduce anime-like output.