edit: this was just to study region-specific prompting

strangely when the guidance is low (2), the result is closer to what was prompted

of course, every model had its own ideas about what a soccer ball may be. I started with “football” but that gave some american results

  • ✧✨🌿Allo🌿✨✧@sh.itjust.worksEnglish
    3·
    3 months ago

    Interestingly for me guidance seems to have less of an affect than resolution. Idk if this is different in different versions of SD but only landscape images reliably makes both balls.

    Square almost never does

    and portrait also very rarely tho slightly more often than square

    • merde alors@sh.itjust.worksOPEnglish
      3·
      3 months ago

      yes, landscape is more loyal to the prompt 👆same configuration with the posted image, except orientation

      thanks