Hacker News Reader

The NSFW thing is just a fig leaf though, it's trivial to disable if you run it locally and it annoys you.

However, it looks like the model wasn't trained on NSFW keywords anyway, which means it is permanently impaired in that regard.

It was trained on LAION 5b, which is chock-full of NSFW material. Search for yourself: https://rom1504.github.io/clip-retrieval/

They filtered a lot of it out, it is a subset of LAION 5b according to undisclosed parameters: https://stability.ai/blog/stable-diffusion-announcement