• fubarx@lemmy.ml
    link
    fedilink
    English
    arrow-up
    3
    ·
    3 months ago

    I tried it with a stock image of red, see-through dice against a solid backdrop. It removed the background, then generated a 2D cutout surface with the dice images projected on the front. Didn’t have time to try any more, but this one result was no good.

    • antipiratgruppen@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      1
      ·
      3 months ago

      Thank you for sharing your brief experience here! It sounds like the opacity of the translucent dice gave the model a hard time, right? Maybe using an opaque object, then changing the opacity in the resulting 3d file would yield a better result? That’s a bit more than 0.5 sec though… and might prove difficult if the input image isn’t your own - though simply editing the input image in various ways could change the way the model handles it, I think!

      • fubarx@lemmy.ml
        link
        fedilink
        English
        arrow-up
        2
        ·
        3 months ago

        Thanks for the tips. Will try again next week. It’s exciting tech, especially when modeling custom objects.

        • antipiratgruppen@lemmy.dbzer0.com
          link
          fedilink
          English
          arrow-up
          1
          ·
          3 months ago

          You’re welcome, and it is indeed!

          Just to be clear - I’m no expert on this. I’m only sharing my thoughts for you or others to try out. I haven’t even experimented with image-to-3d models yet, though I don’t think my assumptions are unreasonable.