Can Adobe Firefly be coached for better accuracy to create science images?

แชร์
ฝัง
  • เผยแพร่เมื่อ 24 ธ.ค. 2024

ความคิดเห็น •

  • @lilyl5492
    @lilyl5492 5 หลายเดือนก่อน +1

    I thought maybe using the kind of logic language of earlier search engine days (e.g. might help it see "water molecule" as one concept), but it doesn't seem to help... or maybe more specific science terms like "H2O" "as atoms of oxygen and hydrogen", but again, similar surreal sci-fi imagery comes out. There may still be a specific prompt language that would get you closer if you can think like the machine rather than like a human? 'circles overlap' representing...?
    I don't think it is set up to recreate 'real things', as we recognise them, possibly because it is inherently a mash up of pixel images detached from context. Maybe someone will build a image generator specifically for science that prioritises meaningful context and integrity as part of the machine learning?

    • @sayostudio
      @sayostudio  5 หลายเดือนก่อน +1

      I definitely agree! In these initial attempts you're so right, I use the language that I know, and that doesn't make the most sense for an AI image generator. I mean, I tried, but I could definitely get more creative about how to do it and get us closer. But then, a water molecule is SIMPLE. If it takes this much work, is it worth it?
      At the end of the day, I'd rather just create the art myself, rather than fighting AI. And let's be honest, even if AI can do many parts of it, I'd still prefer to creat it myself, because that's what I do ;-). But does that mean it's not possible? No.

    • @sayostudio
      @sayostudio  5 หลายเดือนก่อน +1

      And I also agree that it will likely take a targeted image generator that is fed science language and images, with real people teaching it what is and is not correct. That being said, there are other programs out there, like MidJourney, that might already be better at this. I just really don't feel comfortable with how they've built their databases.