• tal@lemmy.today
    link
    fedilink
    English
    arrow-up
    3
    ·
    edit-2
    17 days ago

    Not yet! One thing that AI generated images right now are not so good at is maintaining a consistent portrayal of a character from image to image, which is something you want for illustrating a story.

    You might be able to do something like that with a 3d modeler to pose characters, generate a wireframe, and then feed that wireframe into ControlNet. Or if you have a huge corpus of existing images of a particular character portrayed in a particular way, you could maybe create new images with them in new situations. But without that, it’s hard to go from a text description to many images portrayed in a consistent way. For one image, it works, and for some things, that’s fine. But you’d have a hard time doing, say, a graphic novel that way.

    I suspect that doing something like that is going to require having models that are actually working with 3D internal representations of the world, rather than 2D, at a bare minimum.

    • ✧✨🌿Allo🌿✨✧@sh.itjust.worksOP
      link
      fedilink
      English
      arrow-up
      1
      ·
      edit-2
      16 days ago

      interesting how the ai movie stuff can start with an image of a person and that person be animated in to multiple different activities thru multiple different prompts with the same character. teh future maybe?