Nonsense words can get around DALLE-2’s and Stable Diffusion’s filters

AI Art Generators Can Be Fooled Into Making NSFW Images::Nonsense words can get around DALLE-2’s and Stable Diffusion’s filters

  • lloram239@feddit.deEnglish
    9·
    1 year ago

    Aren’t there NSFW filters after the generation? Bing Image Creator for example will frequently generate images with a borderline-NSFW[1] prompts, but only show you a subset of the four it generated, not all. Some prompts will also be rejected before any generation takes place at all. But I don’t see how this would help you getting through the filter that happens after the generation.

    [1] “borderline-NSFW” really just means anything involving woman or violence, the filter on that thing can be extremely prude and often times a bit nonsensical (e.g. “woman in bikini” is blocked, “woman in 1950 bikini” that’s ok).

    • PersnickityPenguin@lemm.eeEnglish
      2·
      1 year ago

      Yes. Bing will create NSFW content and then flag it and tell you it can’t show it to you because you broke the rules. Lol.