Who will you submit to?

  • j4k3@lemmy.world
    link
    fedilink
    arrow-up
    3
    ·
    7 months ago

    Maybe try stone eyes, creepy face in the negative prompt section. Depending on the model, these are two terms that are part of the age alignment or are close enough to influence it. In general the bad eyes problem seems to be an artifact from how all alignment is over training. These two are related to the underage obfuscation technique used.

    Also, defining a perineum well in the positive, and adding butthole lips, butt hole lips, or even butt hole labia, to the negative prompt can help tremendously. If you positively define something like shinny inside butt hole, it will help too. Most uncensored models do not understand these structures because it is difficult to separate them in the training imagery. Since they are all present in the same image, they are not well defined inside the model. Adding anal cyst to the negative will help with the multi-anus problem. If the model you’re using can handle these terms, this will likely substantially improve the output.

    • Sine Nomen@lemmynsfw.comOP
      link
      fedilink
      English
      arrow-up
      4
      ·
      7 months ago

      Thanks, I’ll keep that in mind. But, uhhh, where do you think I have butthole problems? They don’t even show their butts.

      • j4k3@lemmy.world
        link
        fedilink
        arrow-up
        2
        ·
        7 months ago

        It has to do with the way pussies are messed up in general in SD. They are rather generic and all weirdly wrong. I noticed it in the one shot. The reason they are wrong is usually because it doesn’t know what the difference is between a pussy and an anus. I usually focus on abstract thought. Sorry if I wasn’t clear.

        If you have the ability to run them, lazymixRealAmateur models are really good, especially V4.

          • j4k3@lemmy.world
            link
            fedilink
            arrow-up
            2
            ·
            edit-2
            7 months ago

            Forgot to mention, try DPM, set the steps to 1 and goof around with config steps. Watch how it starts and evolves from 1 config to ~30 and use ultra simple plain text positive prompts with as simple as you can manage negative. Find something you like and set the seed to fixed. This is really powerful for making controlled changes.

            If you step through the config number at ultra fine resolution of like 0.0001 or even less, you will see how the image generation evolves incrementally.

            Lastly, look up neoteny in humans and pedomorphosis on Wikipedia. This is a window into a more scientific description of beauty. Not all the terminology is known to diffusion models, but simply skimming these articles will greatly improve your descriptive awareness.