The question about the legal and moral aspects of training on works of other artists is related, but a different discussion.

  • MrJameGumb@lemmy.world
    link
    fedilink
    English
    arrow-up
    1
    ·
    4 months ago

    They had a rough idea and left it to the AI to make any sense of it and “create” something.

    • Thorny_Insight@lemm.eeOP
      link
      fedilink
      English
      arrow-up
      2
      ·
      4 months ago

      Painters can either splash paint on the canvas or spend months working on a photorealistic masterpiece. There’s absolutely a difference in skill needed for both but to claim the former is not art would also be gatekeeping.

      That argument also disregards the actual difficulty of crafting the perfect prompt to get the AI to output what you want it to. Anyone can create pictures with it but it’s not trivial to get it to create exactly what you want.

      • macniel@feddit.org
        link
        fedilink
        English
        arrow-up
        1
        ·
        4 months ago

        That argument also disregards the actual difficulty of crafting the perfect prompt to get the AI to output what you want it to. Anyone can create pictures with it but it’s not trivial to get it to create exactly what you want.

        I would like to hear what you consider a perfect prompt.

        • Thorny_Insight@lemm.eeOP
          link
          fedilink
          English
          arrow-up
          1
          ·
          4 months ago

          Perfect is probably exaggerated but what I mean is the promp that produces the exact outcome you were looking for. Generative AI can produce very high quality pictures with a simple prompt but if you’re an artist with an exact vision in mind, it not so easy anymore to get AI to produce that for you.

          • macniel@feddit.org
            link
            fedilink
            English
            arrow-up
            1
            ·
            4 months ago

            There isn’t really the exact outcome one was looking for. Not even with a super detailed prompt aided by a control net. LLMs are super imprecise here.