• 16 Posts
  • 10 Comments
Joined 1 year ago
cake
Cake day: June 17th, 2023

help-circle




  • Stable Diffusion on a 4090. I find DDIM sampling is far superior to the default euler_a but it takes fractionally longer. I typically do a 768x512 base image. If it is good then I img2img it to 1.5x size using between 0.4 and 0.6 diffusion. If the results are still good then I upscale using SD Upscale to 2x the size (splits it into pieces) using between 0.2 and 0.3 diffusion.

    At any point if the images sucks I can go back a step and adjust.

    Prompts tend to matter. It was surprising that switching “woman” to “lesbian” for this prompt immediately made the characters objectively less attractive - like visibly older. Adding “ugly” to the negative prompt then corrected that problem while still maintaining subtle things like the haircut.




  • I always think of the scene in silicon valley where Denesh has to look at millions of pics of dicks to train AI…guess it’s hard to find a lot of bored nerds to do that for free.

    There are some impressive LORAs but I haven’t figured out the trick to get in painting to do just that area well. I can make two chicks with dicks…but not my thing. If anyone has a good method for realistic mixed genders in the same scene please share.



  • Surprisingly no - usually they look like clones (as I suspect you have found). Base prompt was “a nude 25yo girl and her 40yo mother with saggy tits standing side by side, ((imperfect skin)), attractive pussy, eyes looking at camera”. I have a preset I use with negatives for disfigurement and cgi and positives for good lighting.

    My typical workflow is to do a smaller image (512x768) and expect them to look like clones. Then I inpaint the people separately with smaller adjustments. I just change the prompt to be “two 40yo women” but only update one of the two people using like 0.6 denoising strength.

    Deliberate_V2 is my model for this. It is pretty awesome.

    SD Upscale to make the higher resolution and use 8x_NMKD-Faces_160000_G as my upscaler.

    Now if only I could figure out how to post the images directly vs a text post with an image embedded lol…