I just tried the glass of wine filled to the brim, and unfortunately, it was able to do it. When I buy something used online, I like to ask for an image with something that ai cant do, so I can be sure the offer is real. Any ideas? There must be something that it still can’t do…

  • DeeBeeDoubleOP
    link
    fedilink
    arrow-up
    2
    arrow-down
    1
    ·
    1 day ago

    That is true. But I feel like asking someone to write half a novel and take a picture of it in front of the object I want to buy is a bit too much. But yeah, text might be something worth looking into.

    • [object Object]@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      1
      ·
      1 day ago

      I just tried generating an image of a ruler obscured by an object, and it seems to struggle with how much should be obscured (e.g. only 5in should be obscured but the number shows that there’s 8 inches behind it).

      I remember how light reflection should converge exactly to a single point. I can’t find a blog post, but it was definitely there too.

      • DeeBeeDoubleOP
        link
        fedilink
        arrow-up
        2
        arrow-down
        1
        ·
        1 day ago

        It works! Three tries in and it cant do it exactly, it always puts the paper close to the number, but never exact. Very nice, thanks!