I am wondering about the implications of deep image generation when it comes to child pornography. It is one of the few kinds of speech that is not protected un

I am wondering about the implications of deep image generation when it comes to child pornography. It is one of the few kinds of speech that is not protected under the First Amendment. How would it be possible to tell an image created with image2text or similar was generated from a model trained on images of child pornography? In this case it would probably still be illegal. Although nobody is harmed from the generation of such an image the courts would probably only care about where the source of the image originated, as with copyright. Or what about using the likeness of a child on the body of a consenting adult for a deepfake? I that case nobody is harmed to begin with.