This is an e621 model, use e621 tags. works well with furry artist tags (inside of fluffyrock) struggles with SD artists.
My original model and still my most loved, stick with this one if you want an overall model and use prompts to push it (say towards realism)
still does adult stuff the best and is better at emulating artist styles/most affected by artists tags.
A push towards realism, the idea behind this one was to have a model that defaults towards realism and then you can use artist tags to make interesting combos.
Suggestions for generation (all links are Civitai links):
A negative e621 inversion is very useful, I highly recommend boring_e621_fluffyrock_v4 as this model is a fluffyrock model at its core
If that one is causing problems then BWU from furtastic is also very good.
1024x768 (or vice versa) is the max I would go for txt2img generation, you're better off with limiting either dimension to 768 to prevent weirdness and then upscaling.
WithoutOrdinary - For their fluffyrock model, this model uses fluffyrock-576-704-832-960-1088-lion-low-lr-e86-offset-noise-e71.
Freek22 - For their Newdawn model, this model uses newdawn_52aurora.
DominoPrincip - For their Opiate model, this model uses opiate_V30PrunedFp16.
A_B - For making stable diffusion fun and the advice on merging ^^ - A_B's models are here.
I recommend taking a look at the above models/creators and giving their stuff a try and a thumbs up.
Removed walls of text from description.
Removed advice on human in negative, I'm finding that it reduces the quality of the picture at least sometimes.