>>2880640The problem is if you describe ANYTHING that could relate to the human body (even parts of the lower body) that could lead the model to generate a human face. You may need to utilize the negative prompt to counter-act this.
Let's take a look at the gallery for the TV head example:
https://files.catbox.moe/9ju4ax.jpeg>score_9, score_8_up, score_7_up, score_6_up, score_5_up, female_focus, <lora:Insert_Head_Here:1> android sign, object_head, h3ad, curvy, fat body, goth, latex, spooky, fishnets, large hanging breasts, low hanging saggy breasts, wide ass, black flip-flops, sexy pose, expressive, stunned, drips of cum on floorNegative:
>watermark, signature, artist name, twitter username, 3d, muscular, bar censor, furry, source_furry, head,mouth,nose,eyes,hat,holding, uglyNamely, "head,mouth,nose,eyes,hat" prevent the model from generating a regular human face. But if the positive prompt had more words describing other body parts like "large hanging breasts, low hanging saggy breasts, wide ass", then that negative prompt would have to be stronger.
And yeah if you're introducing your own loras into the mix that's on you. When I'm working with a new concept I first try things as vanilla as possible; ponydiff, euler a, 30 steps, 1024x1024. Once I get the concept I want consistently, I work in the things that I want from there.