Comment by pousada

15 hours ago

You can’t really because these powerful models are censored. You can create lewd pictures with open models but they aren’t nearly as good or easy to use.

I’ve seen some very high quality NSFW AI video in the last few months. Those models are not far behind and the search and training space for porn is smaller than being able to generate anything

  • > I’ve seen some very high quality NSFW AI video in the last few months. Those models are not far behind and the search and training space for porn is smaller than being able to generate anything.

    Agreed. In my opinion, the primary limitation of the porn models is actually poor labeling of the training set. The company that manages to produce a well-labeled, porn-tuned AI image model is going to absolutely clean up.

    The extractive dark patterns that will emerge from a parasocial chat "AI relationship" that can generate porn images relevant to the chat on the fly will be staggering. Once that proceeds to being able to generate relevant video, all holy hell is going to break loose.

    • > The company that manages to produce a well-labeled, porn-tuned AI image model is going to absolutely clean up.

      For anime/non-photographic content that essentially exists (Pony, then Illustrious, then probably some new-fangled thing by now that I don't even know about), thanks to the meticulously tagged booru image corpus. However, as strong as these models are on matters of anatomy and kinks, they're limited in other ways due to the hugely biased dataset and dependence on tag soup prompts rather than natural language (many find the latter a plus, not a minus, though).

      I haven't heard of any proprietary/cloud-based NSFW model that would be massively better than what's available for free. There are many NSFW-friendly services, but by and large they're just frontends to models trained by other people.

Because models can be used to alter existing images, you can use open and commercial models together in content creation workflows (and also the available findings of open models, and the ability to further tube them very specific used, are quite powerful on their own), so the censorship on the commercial models has a lot less effect on what motivated people can produce than you might think.

I still think, even with that, that like most predictions of AI taking over any content industries, the short-term predictions are overblown.

Doesn't Grok allow users to create lewd content or did they roll that back?

Also, I suspect that we'll soon see the same pattern of open weights models following several months behind frontier in every modality not just text.

It's just too easy for other labs to produce synthetic training data from the frontier models and then mimic their behavior. They'll never be as good, but they will certainly be good enough.

Just a matter of time and open models will get there. Not once have we seen a moat across the model spectrums.