As this was a more elaborate model, it was more difficult than expected. I had to try several checkpoints to make the face not so far away from the original. The diffusion models were reluctant to generate the more elongated shape of the breast, so I had to model everything in 3d using Makehuman. I never cease to be amazed by the diffusion models: it did a beautiful job on the diffraction of the body underwater.

  • ~~Nudermeisters~~@lemmynsfw.com
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    1 year ago

    so I had to model everything in 3d using Makehuman

    As in, for use with ControlNet depthmap? I remember using that program way back in the day for Poser haha. this image heavily reminds me of Poser. try adding “poserpro” and “daz” or “daz studio” in to see if the models ever got trained on some renders. I bet they were.

    • halftimecoder@lemmynsfw.comOP
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 year ago

      I use ControlNet tile, normal map and inpaint, depending on the task. Usually, I create many variations using X/Y/Z script testing different CFG and denoise values, and in GIMP I create mockups that are feed into SD.

      About Poser, could it be? It would be interesting to know how they create the original art.

      • ~~Nudermeisters~~@lemmynsfw.com
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        1 year ago

        1987 was way too early for Poser I think. I meant that your art here, looks like an early Poser render, so adding those tags, may assist with capturing this art style some more. After some digging, Sierra’s “AGI” Adventure Game Interpreter was used for #1, SCI (Sierra Creative Interpreter) was the spiritual successor to AGI, and weas used for some of the other LL games. I dare you to train a LoRA on that