Description:
Kaylee (not her real name) likes to play guitar, raise bunnies and talk about philosophy. She doesn’t eat meat, is against animal experimentation and is currently studying to become a trial lawyer. (She sees herself as a person who likes to argue using logic and reasoning, so she wants to use her skills to defend those in need.)
Procedure:
I got the first image by accident. It was part of a bunch using wildcards. The prompt was:
close up of a normal girl from a porn magazine \_\_female\_names\_\_ \_\_female\_hairstyles\_\_ \_\_hair\_color\_\_ \_\_female\_clothing\_\_ {back|side|front} view (nsfw) {surprised|smiling|laughing}
Then, I sent that image to inpaint to cover some details as the picture was showing too much of her. (She is kind of shy about her body and desn’t like to expose too much of it. She actually refuses to accept she is pretty.) Then, I used that image as reference loading it in the ControlNet input in the txt2img tab to generate new ones, using values like: 1,0,1, 0.5,0,0.5, 1,0,0.75 or similar, sometimes loading also some of the newly generated images. I also used OpenPose and SoftEdge and maybe some other in order to get variations. In some occasions, I had to sent the low resolution image to the ControlNet input to make an inpaint harmonious of the image in high resolution, using values usually of 1,0,1. All the other settings remain the same. This can make the image to have more details and contrast, which may or may not be desirable.
Negative prompt:
(((large breasts))), cartoon, painting, illustration, (worst quality, low quality, normal quality:2), ((disfigured)), ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), poorly drawn hands
Parameters:
Steps: 12, Sampler: UniPC, CFG scale: 3, Size: 512×853, Model: photon\_v1, Denoising strength: 0.35, Clip skip: 1|2, Token merging ratio: 0.5, Token merging ratio hr: 0.5, RNG: CPU
Hires upscale: 3.15, Hires steps: 12, Hires upscaler: 4x-UltraSharp, Version: v1.5.0
Notes:
The original image uses a Clip skip value of 1, the others ones use 2. The first images are a bit too similar (I made more, but couldn’t get rid of all of them), but as some form of compensation the last ones are the most revealing.
