My SDXL model but speedy.
Trained/Merged but mostly Addift training using ~300k imgs.
Already has all possible femdom-loras and pov-loras like low-angle view.

Negative-embeds: Not necessary / Harmful even in some cases (doing a double-negative-effect).

You should be fine if you only use your own stuff. (no need negative loras)
Or add:
add in negative:
'(blurry:1.2), (poorly drawn:1.3), (worst quality:1.5), (bad artist:1.1), (unclear:1.2)'


7 steps.
2.0 cfg
dpmpp_2s_ancestral_cfg_pp
karras

Resolutions:
768x1344 (Best)
1024x1024 (Second best)

Using:

NAI Vpred
NAI Danbooru prompting