Hello everyone,

I am just starting to get a hang of this stuff, and am interested in the challenge of any requests from the community.

Please provide me with:

  • Subject/Person details
  • Pose/Composition
  • Aesthetic/Style
  • Location/Background
  • Whatever else comes to mind
  • korewafap@lemmynsfw.com
    link
    fedilink
    English
    arrow-up
    5
    ·
    1 year ago

    Half Asian half Spanish women in her 30s talking a selfie in a messy bedroom wearing only black lingerie and the bed is visible. Hair has a balayage hair color. Make it look like it came from an iPhone

    • justinreplica1@lemmynsfw.comOP
      link
      fedilink
      English
      arrow-up
      4
      ·
      edit-2
      1 year ago

      This one was humbling. I didn’t end up getting the exact result I wanted, but these are the best of the bunch. I’m a big fan of the second one.

      Hope you enjoy.

      • korewafap@lemmynsfw.com
        link
        fedilink
        English
        arrow-up
        4
        ·
        1 year ago

        These are great. I like the last one the most.

        I actually requested what I did because I’m having a hard time too. I have a face Lora trained but can’t get it to look convincing.

        • justinreplica1@lemmynsfw.comOP
          link
          fedilink
          English
          arrow-up
          1
          ·
          edit-2
          1 year ago

          I have found consistent character models to be hard to get right, and really sensitive to certain configurations. Try with different checkpoints, CFG, clip skip. Also some loras respond better with additional trigger words - maybe the model you have is documented.

          Most of what I have learned has been from looking at examples and then using the PNG info to attempt to recreate it exactly with all the same model SHAs