Not in the US, but in Canada and many European countries, I believe it's illegal.
> We cannot be at a point as humanity where text is illegal. Right?
Even in the US, much text is illegal in certain contexts. Think false advertising, written plausible threats, etc.
I have to wonder about the legal implications of an AttnGAN or word-to-image type model trained on nothing but illegal pornography. Or one where the ratio of legal to illegal content the model was trained on is unknown and impossible to determine. There are some jurisdictions that consider synthetic images or text to be a victimless crime. At what point does the line become crossed and the images start becoming recognizable and realistic enough to be illegal at first glance under any usual circumstance, but are actually generated from nothing? How would people be able to prove the origin and veracity of such images for the purpose of submitting evidence of a crime?