Trying out a different workflow this time. Trying to solve the issue of maintaining details when increasing the resolution. I find the higher resolution I go, the more denoised the image becomes. The workflow is something like this: PAG (Perturbed Attention Guidance) => Kohya Deep Shrink + AYS => TCD Sampling => Ultimate Upscaler => Inpaint + manual adjustments => Img2Img w/ Sonar sampling + TTPlanet Tiling ControlNet => Gradient Removal => Snapseed/Procreate.
Prompt
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_anime, (small petite girl, loli:1.1), embarrassed, blush, wavy mouth, open mouth, pale_skin, in a dark apartment, lifting own swimwear, bikini-lift, exposed breasts, looking at the viewer, japanese idol, ponytail, platinum_blonde, diffused window light, whispering_bath_tub, (orange colorwash:1.2), (side profile:0.7), steam
Negative prompt
score_6, score_5, score_4, censored, logo, icon, simple background, white background
"Eh?? W-Well if you insist..."
Trying out a different workflow this time. Trying to solve the issue of maintaining details when increasing the resolution. I find the higher resolution I go, the more denoised the image becomes. The workflow is something like this: PAG (Perturbed Attention Guidance) => Kohya Deep Shrink + AYS => TCD Sampling => Ultimate Upscaler => Inpaint + manual adjustments => Img2Img w/ Sonar sampling + TTPlanet Tiling ControlNet => Gradient Removal => Snapseed/Procreate.