Young professionals are turning to AI to create headshots. But there are catches::undefined

  • drekly@lemmy.world
    link
    fedilink
    English
    arrow-up
    24
    ·
    11 months ago

    But instead of learning stable diffusion and training their own model for better results for free, they’re paying a service for 100 images that likely won’t be well trained.

    • 8ender@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      1
      ·
      10 months ago

      I love this take because it’s the modern ML version of shaming someone for not brewing their own beer

      • drekly@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        10 months ago

        I’m speaking in the context of the article which posits that AI is flawed.

        I’m arguing that it’s not flawed if you do it right, but the services that provide cheap AI headshots are lacklustre.

        I don’t expect everyone to learn it, but I also don’t think you should say it’s flawed unless you’ve tried to do it properly.

  • bbbbb@lemmy.world
    link
    fedilink
    English
    arrow-up
    6
    ·
    edit-2
    11 months ago

    Yeah, my wife tried one of these services and got terrible results. They really do give better results to certain people’s faces depending on the training set

    • Yuki@kutsuya.dev
      link
      fedilink
      English
      arrow-up
      3
      ·
      11 months ago

      Basically, yes. I train models as a hobby and this is true since it depends on what it’s trained on.

      Let’s say I train a model purely on Caucasian women, it’ll be hard to get an Asian male out of the model since it’s not trained on it.

      Same goes for other features like hair color, eyes, nose, etc.

      • bbbbb@lemmy.world
        link
        fedilink
        English
        arrow-up
        4
        ·
        edit-2
        11 months ago

        Yeah, makes sense, the people I saw online with “successful” results were primarily white men, which I would imagine would pop up pretty frequently in a training set. The bias also feeds into the echo chamber of AI hype too in my opinion