Experiments with 2:1 aspect ratio

  • Cavendish@lemmynsfw.com
    link
    fedilink
    English
    arrow-up
    6
    ·
    1 year ago

    You can get a lot of interesting pose variety by messing with the aspect ratio. See also regional prompting to carve out spaces within the larger frame. I find putting head/hair/face prompts in their own region, then scaling that region, to be extremely effective in controlling close-up to wide shot framing.

    • Wrench Wizard@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 year ago

      Amazes me that AI can do all this stuff yet can’t understand basic common language instructions ie: character here, in this pose, with this background etc.

      Seems like at this point we’d just be writing what we want to see in… layman’s terms

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

        You have to use a conversational AI to talk to the other AI to generate the image. It’s the AI equivalent of running sentences through Google translate multiple times.

    • AppleJuiceGoblin@lemmy.worldOP
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      Wow, this seems like a cheat code. Thanks! I’m curious. How are you scaling a specific region? Is that just messing with the ratio?

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

        Exactly. Think of a portrait orientation image: top 25% sky, next 25% head, bottom 50% torso. Will come out way different than top 60% head, bottom 40% chest. Using keywords like “closeup, medium shot, cowboy shot” are less effective for me, but that’s what you see in lots of tutorial posts for controlling composition through prompting alone. You can even go crazy with the positioning. Portrait photo split vertically, with head in the left column, body in the right column, will make them lean over or arched back, etc.

        • ~~Nudermeisters~~@lemmynsfw.com
          link
          fedilink
          English
          arrow-up
          3
          ·
          1 year ago

          tonight i’ve noticed that varying a subject’s age, and nothing else, using “xx year old” format, can reliably change a lot of other interesting things. Try it out in a prompt matrix some time. i found 18-24 were more likely to be nude, and when “reading” would have a small pamphlet. 25-29 would get sexy lingerie and paperback book. 30-36 would get underwear and books that appeared to be science books. Obviously this was all situationally dependant, but i found it interesting anyway.

        • AppleJuiceGoblin@lemmy.worldOP
          link
          fedilink
          English
          arrow-up
          2
          ·
          1 year ago

          Very cool idea! Maybe that’s better than loras even. I’m already having fun using different dynamic prompts with wildcards to get surreal scenery. Thanks for the info!