• PM_Your_Nudes_Please@lemmy.world
    link
    fedilink
    English
    arrow-up
    4
    arrow-down
    46
    ·
    1 year ago

    While I don’t disagree with the initial premise, image AI requires training images.

    I suppose technically you could use hyper-realistic CGI CSAM, and then it could potentially be a “victimless” crime. But the chances of an AI being trained solely on CGI are basically non-existent. Photorealistic CGI is tough and takes a lot of time and skill to create from scratch. There are people whose entire careers are built upon photorealism, and their services aren’t cheap. And you’d probably need a team of artists (not just one artist, because the AI will inevitably end up learning whatever their “style” is and nothing more,) who are both capable and willing to create said images. The chances of all of those pieces falling into place are damned near 0.

    Maybe you could supplement the CGI with young-looking pornstar material? There are plenty of pornstars whose entire schtick is looking young. But they definitely don’t look like children because the proportions are obviously all wrong; Children have larger heads compared to their bodies, for example. That’s not something that an adult actress can emulate simply by being flat chested. So these supplemental images could just as easily end up polluting (for lack of a better word) your AI’s training, because it would just learn to spit out images of flat chested adult women.

    • db0@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      35
      arrow-down
      2
      ·
      1 year ago

      Generative Ai is perfectly capable of combining concepts. Teach it how do today do photorealistic underage and photorealistic porn and or can combine them together to make csam without ever being trained on actual csam